Research Article A Data Hiding Technique to Synchronously ...downloads.hindawi.com/journals/bmri/2015/514087.pdfData hiding is another alternative for video-waveform synchronization.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Research ArticleA Data Hiding Technique to Synchronously Embed PhysiologicalSignals in H264AVC Encoded Video for Medicine Healthcare
Raul Pentildea Alfonso Aacutevila David Muntildeoz and Juan Lavariega
Tecnologico de Monterrey 64849 Monterrey NL Mexico
Correspondence should be addressed to Raul Pena raulportegagmailcom
Received 2 January 2015 Accepted 24 February 2015
Academic Editor Cheng-Hong Yang
Copyright copy 2015 Raul Pena et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
The recognition of clinical manifestations in both video images and physiological-signal waveforms is an important aid to improvethe safety and effectiveness in medical care Physicians can rely on video-waveform (VW) observations to recognize difficult-to-spot signs and symptoms The VW observations can also reduce the number of false positive incidents and expand therecognition coverage to abnormal health conditions The synchronization between the video images and the physiological-signalwaveforms is fundamental for the successful recognition of the clinical manifestations The use of conventional equipment tosynchronously acquire and display the video-waveform information involves complex tasks such as the video capturecompressionthe acquisitioncompression of each physiological signal and the video-waveform synchronization based on timestamps Thispaper introduces a data hiding technique capable of both enabling embedding channels and synchronously hiding samplesof physiological signals into encoded video sequences Our data hiding technique offers large data capacity and simplifies thecomplexity of the video-waveform acquisition and reproduction The experimental results revealed successful embedding and fullrestoration of signalrsquos samples Our results also demonstrated a small distortion in the video objective quality a small increment inbit-rate and embedded cost savings of minus26196 for high and medium motion video sequences
1 Introduction
Video technology continues to improve the safety and effect-iveness in healthcare Today physicians and engineers rely onrigorous video-based studies to improve medical practicesand proceduresThese studies are necessary for the identifica-tion of clinical manifestations in patients and for the reduc-tion of errors duringmedical procedures Extending the video-based studies to incorporate the analysis of physiological-signal waveforms further enhanced the recognition of clinicalmanifestations and the reduction of false positive casesPhysicians can rely on simultaneous video-waveform obser-vations to recognize difficult-to-spot signs and symptomsThese observations can also expand the recognition coverageto abnormal health conditions
The synchronization between the video images and thephysiological-signal waveforms is fundamental for enhancedrecognition of the clinical manifestations The identificationof signs and symptoms invisible during specific diagnosisis possible with synchronized video-waveform observations
Physicians are able to diagnose seizures in neonates after cre-ating and analyzing a permanent record [1] The permanentrecord needed for this diagnosis contains synchronized videoand electroencephalographic (EEG) waveforms
Three commonly used techniques are suitable for video-signal synchronization [2] The timestamp-based techniqueis themost common alternative for video-audio synchroniza-tion This technique inserts time codes at each signal streamThese time codes are also useful for future browsing storageand reproduction of permanent records [1] However thetimestamp-based synchronization involves complex taskssuch as the capturecompression of the video signal the acqui-sitioncompression of each physiological signal the insertionof timestamps into individual streams and the use of special-ized software for stream synchronization [2]
The second technique relies on synchronization marksThe technique sends synchronization marks from a transmi-ssion node The disadvantage of this technique is the needof an additional assistant communication channel to trans-mit the synchronization mark The third technique is
Hindawi Publishing CorporationBioMed Research InternationalVolume 2015 Article ID 514087 10 pageshttpdxdoiorg1011552015514087
2 BioMed Research International
Timestampsmultiplexing
marks
Videoencoder Synchronization process
Transmissionstorage
0 1 2 3
Time (s)
0 1 2 3
Time (s)
09998
09989
09985
09973
09944
09913
09998
09989
09985
09973
09944
09913
(a)
Videoencoder withdata hiding
Transmissionstorage
09998
09989
09985
09973
09944
09913
099980998909985099730994409913
0 1 2 3
Time (s)
0 1 2 3Time (s)
(b)
Figure 1 A comparison between (a) conventional synchronization process and (b) our proposed approach
multiplexing This technique maintains the correlationamong the media streams during the transmission processHowever multiplexing-based synchronization usually resultsin a loss of agility and integrality
Data hiding is another alternative for video-waveformsynchronization Data hiding has the goal of embeddinginformation into encoded video sequences with a minimumamount of perceivable degradation [3] The embedded infor-mation can be text pictures or physiological-signal samplesData hiding synchronizes the video and signals after hidingeach physiological sample at its corresponding video framein time
This paper introduces an improved data hiding techniquewith a larger data-hiding capacity in the context of medicalhealthcareOur data hiding technique synchronously embedsphysiological-signal samples intoH264AVC-encoded videosequencesThe implementation of data hiding is simpler thanother synchronization techniquesThe data-hiding techniquerequires only one encoder to process video and signalrsquossamples Data hiding also offers the advantage of a uniquecommunication channel for video-audio transmission andrequires no complex tasks related to timestamps synchro-nization marks or multiplexing [4 5] Figure 1 illustrates acomparison between the commonly used synchronizationprocesses and our synchronization approach based on datahiding Other important advantage is that our techniqueis strongly related to trends in secure handling of medicaldata during signal transmission Data hiding makes possible
the secure transmission of patientrsquos information over theinternet Important features for secure transmission of per-sonal information are authentication integrity and confiden-tiality [6]
2 Background
21 Data Hiding Techniques Existing data hiding techniquesare able to hide information in the video frames duringthe video encoding process In [3] authors proposed a datahiding scheme based on the macroblockrsquos size needed bythe H264AVC interprediction process The scheme is ableto hide two bits per macroblock and requires the followingpartitions types 16times 16 16times 8 8times 16 and 8times 8 The schemeloses no hidden data and may result in bit-rate increments
In [7] the data hiding scheme relies on constrainsassociated with the H264AVC interintraprediction modesIn the interprediction mode the scheme hides 0 bits at theinterprediction mode using the block sizes 16 times 8 8 times 168 times 4 and 4 times 8The scheme also hides a 1 bit using the blocksizes 16 times 16 8 times 8 and 4 times 4 In the intraprediction modethe scheme hides 0 bits using the block sizes 16times16 and 4times4Hiding a 1-bit value requires the 8 times 8 block size The schemehas minimum impact on the video quality and controls thedistortion degradation by hiding no data in 4 times 4 blocks
In [8] the data hiding scheme embeds information dur-ing context-based adaptive variable length coding (CAVLC)The scheme hides one bit using the trailing ones parity of
BioMed Research International 3
Table 1 Qualitative metrics and statistics of the data hiding (DH) schemes reviewed
Reference DH schemes DH capacity Max PSNR119884distortion Max bit-rate distortion
[3] Forcing block-type partitions 2 bits per MB (interframes) minus14 dB (40 kbps) 7000 bps[7] Grouping block-type partitions 1 bit per MB (interintraframes) minus009 dB 465[8] Parity of trailing ones CAVLC 1 bit per MB (intraframes) minus257 dB 0039[9] Quarter-pixel search positions 1ndash4 bits per MB (interframes) minus003 db 131[10] Last nonzero coefficient parity 1 bit per MB (interframes) Approx minus001 dB 1200 bps[11] Motion vectors and mode sel 1 bit per frame (interframes) 003 dB 098
the CAVLC code-word The CAVLC process results in mod-erate visual degradation and maintains the overall size of thevideo stream
In [9] the proposed technique exploited quarter-pixelmotion estimation process to hide dataThe schemehides onebit by modulating the best search points of a subblock Therate-distortion cost is introduced to reduce both the impacton the video quality and the increment in the bit-rate aftersearch point adjustments The hiding capacity is dependenton the content of the video sequence
In [10] the authors proposed a data hiding scheme basedon an adaptive method The method hides one bit using thelast nonzero coefficient parity after quantization of a 4 times 4luma block The scheme relies on an adaptive rather thana fixed point for data embedding The scheme results ina proportionally direct behavior between the bit-rate andcapacity size and between the bit-rate difference and theamount of embedded data
In [11] the proposed data hiding scheme relies onmotionvectors and mode selection The scheme only hides one bitper frame This scheme embeds data using the macroblocksearch regions with a left area restriction for a 1 bit and a rightarea restriction for 0 bits
In [4 5] the authors demonstrated that existing datahiding schemes in [8 10] successfully embedded audio intoencoded video sequences with minor impact on video imagequality and bit-rate
The desirable features of a data hiding technique suitablefor embedding physiological signals into encoded videosequences are a large data hiding capacity a low impact invideo quality and aminor effect in bit-rate of the video Table 1compares the data hiding schemes previously reviewed con-sidering three metrics hiding data capacity the maximumPSNR
119884(objective video quality) and bit-rate distortion
These schemes offer relatively low values for the threemetricsTherefore the main limitation resides in the data hidingcapacity In [9] the quarter-pixel motion estimation schemeoffers the highest data hiding capacity using an 8 times 8partitionThis scheme also offers very low PSNR
119884distortion and less
than 132 of bit-rate distortion Our proposed techniqueextends the quarter-pixelmotion estimation scheme to satisfythe data hiding capacity needs and to ensure the low PSNR
119884
and bit-rate distortions
22 Motion Estimation in H264AVC Motion estimation(ME) is an important element in the H264AVC interpre-diction process For a given frame the ME goal is to find
Table 2 Examples of medical applications
Reference Medical application Phys signals[1] Silent neonatal seizures EEG
[14] Patient safety in anesthesiaoperating rooms Vital signs
the best predictions for both levels macroblock (MB) selec-tion and motion vector (MV) estimation A MB is an arrayof 16 times 16 pixels The MB selection process assumes thepartitioning illustrated in Figure 2 Each partition containsa MV value Equation (1) shows how to select the bestblock partition by calculating the Lagrangian rate distortion(119869mode) optimization In this equation120582mode is the Lagrangianmultiplier SSD is the sum of the squared difference betweenthe original and the reconstructed block and119877 is the numberof bits of MB parameters such as quantization parameterheader motion vectors and residue coefficients
119869mode = SSD + (120582mode) (119877) (1)
The motion estimation process computes motion vectorsfor each macroblock partition found in each video frameAt a given frame the ME process searches for the new MBposition of each MB located in the reference frame The MEprocess calculates motion vectors based on these new MBpositions and encodes these vectors in the encoded framesFigure 2 illustrates the three ME stages to compute a MVThe first ME stage identifies the best MB position at theinteger-pixel mesh The second ME stage identifies the bestMB position at the half-pixel mesh based on the best integer-pixel position The third ME stage identifies the best MBposition at the quarter-pixelmesh based on the best half-pixelposition The selected position becomes the final MV value
23 Application Examples of Video-Based Medical Care Thissection presents additional examples of medical applica-tions related to the simultaneous observation of video andphysiological-signal waveforms Table 2 shows the names ofthe applications and the specific physiological signals neededfor simultaneous correlation with the video In [1] physicianstake advantage of synchronized EEG recordings with video to
4 BioMed Research International
Macroblock partitions Motion estimation (ME)
16 times 16
8 times 16
16 times 8
8 times 8
8 times 8 8 times 4
4 times 44 times 8
Stage 1Stage 2
Stage 3Selected
Submacroblock partitions
Figure 2 Macroblock partitions and search points in motion estimation
correlate clinical manifestations such as lip smacking fixingof eyeballs and cyclic leg movements
In [12] physicians take advantage of synchronized digitalvideo recordings to identify nocturnal breathing anomaliesusually undetected by standard polysomnography Successfulidentification of these anomalies requires the correlationamong EEG recordings oxygen saturation (SpO
2) endtidal
CO2level in-video leg movement and in-video rapid eye
movement (REM)In [13] the purpose is to assess the validity of a new
physical-activity monitor in the context of congestive heartfailure This monitor utilizes body-fixed accelerometers todistinguish among activities such as body postures sit-ting standing normal walking stairs walking cycling andwheelchair driving These in-video activities are correlatedwith the accelerometer waveforms to assess the correctoperation of the activity monitor
In [14] clinical investigators perform rigorous studies toenhance patient safety in operating rooms The investigatorsfirst elaborate a permanent feedback record containing the in-video health delivery process vital signs and other signalsThen the investigators reproduce this permanent recordto observe and to assess the health delivery process Thesynchronization of the video and the physiological-signalwaveforms is fundamental for the identification of factorsresulting in adverse events
3 The Proposed Data Hiding Technique
Our proposed technique hides streams of data samplesinto encoded video sequences The implementation of ourtechnique is a set of software routines added to the original
0 1 14 15 16 17 30 31
Time
EEG
Videoframes
middot middot middot
middot middot middot
middot middot middot
middot middot middot
fr = 30 framess
rEEG = 256 sampless
Figure 3 Synchronization process
H264AVC codec Figure 3 illustrates the synchronizationbetween the video and the physiological-signal waveformsof EEG samples Our technique synchronizes video andEEG signals by hiding samples of these signals at theircorresponding frame in time
31 Encoding Process in Our Data Hiding Technique Ourtechnique modifies the H264AVC interprediction processto hide the samples of the physiological signals Our tech-nique relies upon the quarter-pixel motion estimation pro-cess part of the H264AVC encoder to hide these samplesinto encoded video frames Our technique divides the searchpoints into four groups to improve the data hiding capacity ofthe original technique [9] Equation (2) presents the proposedpartitioning of search points and the binary assignment foreach partition The expression in (3) indicates how to select
BioMed Research International 5
Video
Transform PS sample
Mapping ME
MEno
sample
Motioncompensation
Current video frame
Video
Previouslycoded frames
Yes
No
force
frames
frames
14 pixel SP
Intra iquestallsamples
Intracoding Intercoding
intercoding
Figure 4 The block diagram of our proposed data hiding technique based on H264AVC encoder
Algorithm Embed a sampleInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(6) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(7) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(8) case 11 Quarter-Pixel Search Position = 0(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 1 Algorithm of our data hiding encoding technique
the search point in each group with the minimum distortioncost In (3) 119869119894 is the Lagrangian rate distortion parameter
Our proposed technique illustrated in Figure 4 repeatsthe gray blocks until no samples are available The blocksin gray are the additional routines needed to implementour proposed data hiding algorithm The alternate pathexecutes the original H264AVCmotion estimation processAlgorithm 1 presents details of our data hiding algorithmfor encoding Our technique hides the signal samples in themotion vectors of each block partition located at a frameEach motion vectors hides two bits of a signal sample Ourtechnique hides no samples in the block types I4MB andI16MB due to their association with intraframe prediction
Our technique is also unable to hide samples into PSKIPblocks due to the lack of motion vectors
Our technique also incorporates an approach to over-come the data-hiding capacity limitation found in low-motion video sequences PSKIP blocks are the most com-mon block partition found in encoded low-motion videosequences A large number of PSKIP blocks limit the data hid-ing capacity of our technique due to few number of motionvectors found in the low-motion video sequence Thereforeour proposed technique forces the H264AVC encoder toreplace the PSKIP block by a P16 times 16 block partitionThis PSKIP replacement adds a motion vector to the datahiding capacity of our techniqueThePSKIP replacement alsocontributes to maintaining the synchronization between theencoded video and the physiological signals However thisreplacement may also result in an increment in the bit-rate ofthe sequence Algorithm 2 presents details of the algorithmfor low-motion sequences
32 Decoding Process of Our Data Hiding Technique Ourdata hiding technique extracts the samples of the EEG signalat the same time conventional video H264AVC decoding
6 BioMed Research International
Algorithm Embed a sample in low-motion sequencesInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) if PSKIP partition do(4) curblock = P16 times 16 partition type(5) switchMapping do(6) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(7) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(8) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(9) case 11 Quarter-Pixel Search Position = 0(10) end switch(11) end ifrarr until complete all partitions of the current macroblock(12) end forrarr until complete all samples
Algorithm 2 Algorithm of our data hiding encoding technique for low motion sequences
Referencevideo
frames
Read MV ofMB partitions
Extractbit pairs
Physiologicalsignaloutput
Video
Encoded videosequence
No
YesConcatenated
value
iquestallsamples
Figure 5 Sample extraction during the decoding process
process takes placeThis is an important feature at the time todo synchronized playback of both video and physiological-signal waveforms The sample extraction algorithm in ourtechnique catches and gathers sample data This extractionalgorithm illustrated in Figure 5 is repeated as many timesas needed to extract all the samples embedded in the videosequence To do this our data hiding technique interacts withthe H264AVC decoding process
The routine reads the motion vector (MV) of every mac-roblock partition and inputs the MV
119909and MV
119910components
into (4) to identify a binary combinationThe |MV|2 termsoutput the quarter-pixel search point values The routineobtains the bit-pair samples after identifying the 1198661 to 1198664values mapped into (2)
type isin
1198661 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
1198662 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198663 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198664 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
(4)
Algorithm 3 presents details of our proposed decodingalgorithm This decoding process extracts the physiologicalsamples from the encoded video sequences
4 Results
Theexperimental setup illustrated in Figure 6 included a PCan EEG database and a set of seven video test sequencesThe EEG samples were extracted from the CHB-MIT ScalpEEGDatabaseThis database is located at the PhysioBankdig-ital recordings (httpwwwphysionetorg)The experimentalsetup included 6 signal electrodes at 256 samples per secondand 12-bit sample resolution [1] The EEG samples generatedin one second were embedded into the first 30 frames of eachtest video sequence to establish synchronization
The implementation of encoding and decoding processesneeded the modification of the JM reference software version161 Table 3 shows the JM configuration A program wasdeveloped to provide and convert samples from the databaseto the encoder
The video test sequences had a CIF (352times288) resolutionand a 420 YUV format The name of the test sequences are
BioMed Research International 7
Algorithm Extract a sampleInput Embedded sample in a synchronous frameResult Data of the sample in bit-pairs according to (2)(1) for ExtractSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case (|MV
119909|2 = 1) amp (|MV
119910|2 = 0) Bit-pair = 00
(6) case (|MV119909|2 = 0) amp (|MV
119910|2 = a) Bit-pair = 01
(7) case (|MV119909|2 = 1) amp (|MV
119910|2 = 1) Bit-pair = 10
(8) case (|MV119909|2 = 0) amp (|MV
119910|2 = 0) Bit-pair = 11
(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 3 Algorithm of our data hiding decoding technique
Experimental setup
Video testsequences
Neonatal
EEG
samples
Storage
Transmission
ModifiedJM reference
H264AVC
09998
09989
09985
09973
09944
09913
0 1 2 3
Time (s)
Figure 6 Experimental setup for our synchronization data hiding scheme
Table 3 Configuration parameters for JM reference software
Parameter ResolutionProfile BaselineFrames 30Motion estimation algorithm Full searchRD optimization and rate control Disabled8 times 8 subblocks DisabledNumber of reference frames 1Quantization parameter (QP) 28
akiyo bridge-far carphone football foreman mobile andneonatal Neonatal is not considered a standard video testsequence Neonatal was introduced to match the context ofthe application example related to EEG seizures on neonatesThe selected coding structure of the bit stream is ldquoIPPP rdquo tohave an intraframe encoded in the first frame and interframesencoded in the remaining frames
Metrics to evaluate the effectiveness of our proposed tech-nique are video objective quality bit-rate difference embed-ding cost and perceptual quality of the image The peak
signal-to-noise ratio (PSNR) illustrated in (5) is an objectivequality metric to report video image degradation119872 and 119873are the height and the width of the video frame respectively119868119894119895 and 1198681198941198951015840 represent the original pixel and the processedpixel respectively [5] The PSNR
119884diff metric illustrated in(6) indicates how the luma (119884) samples impact the videoquality after embedding the physiological samples PSNR1015840
119884
represents the impact of luma samples generated by embed-ding the samples and PSNR
119884represents the impact of luma
samples generated with the original H264AVC encoder
PSNR = 10 logmax (1198681198941198952)
(1119872)sum
119872
119894sum
119873
119895[119868119894119895 minus 119868119894119895
1015840]
2 (5)
PSNR119884diff = PSNR
1015840
119884minus PSNR
119884
(6)
Equation (7) shows how to calculate the change of bit-rate (BRI) 119877 is the original bit-rate and 1198771015840 is the embeddingsamples bit-rate Equation (8) shows how to estimate theembedded cost Oe Ov is the data volume generated by theoriginal video coder DHe is the data volume generated byembedding the samples and EEGe is the embedded datavolume [7] The EEGe term refers to the data coming from
8 BioMed Research International
Table 4 Experimental results for seven video test sequences
Figure 7 PSNR luma difference values for neonatal video sequence
Football Mobile Foreman Carphone Neonatal Akiyo Bridge-far
Video test sequences
High motionMedium motionLow motion
00005
0010015
0020025
0030035
minus001minus0005PS
NR Y
diffe
renc
e (dB
)
Figure 8 PSNR luma difference for seven video test sequences
the EEG signals The perceptual quality provides an estima-tion of subjective quality of the image obtained by visualinspection
BRI = 1198771015840minus 119877
119877
times 100 (7)
Oe = DHe minusOv minus EEGeOv + EEGe
times 100 (8)
Figure 7 compares the PSNR119884difference between the
original and the embedding data encoding processes Thegraph shows a very small difference between PSNR
119884values
for the first 30 frames of the neonatal video sequence Thelargest increment in quality was 0187 dB This incrementappeared in the 29th frame The largest decrement in qualitywas minus0039 dB This decrement appeared in the 8th frame
Figure 8 demonstrates that the embedding process hasminor effect on the video objective quality of seven testsequences The 119910-axis represents the PSNR luma differencevalues expressed in decibels The positive values above aver-age 0011 dB and 0029 dB represent a very small improve-ment in video qualityThe negative values from minus0002 dB to
05
10152025303540
Bit-r
ate d
iffer
ence
()
Carphone Neonatal Akiyo Bridge-farFootball Mobile Foreman
Video test sequences
High motionMedium motionLow motion
Figure 9 Change of bit-rate for seven video test sequences
minus0007 represent a small degradation in video quality Thesedegradations and improvements are due to the lack of bit-rateconstraint
Embedding the EEG data into the video sequencesproduced increments of bit-rate for the seven video testsequences as illustrated in Figure 9 Our experimental setupincluded high motion medium motion and low motionvideo test sequences The amount motion of the neonatalvideo sequence was considered between medium and lowThe graph presents the neonatal sequence with gray colorto indicate that it is not a standard video test sequence Forthe high and medium motion sequences there are small bit-rate increments The bit-rate increments ranged from 045to 134 For the neonatal and low motion sequences thePSKIP block replacement occurred and resulted in bigger bit-rate incrementsThe bit-rate increments ranged from 56 to3658 However the changes in the bit-rate had no effect onthe video-waveform synchronization
Table 4 shows the results for the seven test sequences interms of embedded capacity modified macroblocks mod-ified motion vectors PSNR
119884difference bit-rate and the
embedded cost For the high medium motion and neonatalsequences the embedded cost ranged from minus01273 tominus26196 representing savings in data volume For the lowmotion sequences the embedded cost had an incrementin 158726 and 193017 These results indicate that ourdata hiding technique offers both an adequate efficiency forvideo-signal transmission and savings in the storage of highmedium and neonatal sequences
The inspection of subject quality indicated minimumvisual artifacts or distortion between original and data-hiding
BioMed Research International 9
(a) (b)
(c)
Figure 10 10th frame of neonatal video sequence (a) Original image (b) image using video original coding and (c) image with EEG data
images Figure 10 illustrates the perceptual quality of thevideo frames for the 10th frame of the neonatal sequenceThus data hiding generates no significant difference inquality from a human eye perception
Our technique will offer an adequate performance in thecontext of the application examples presented in Section 23The video sequences of these applications exhibit sufficientamount of motion In the neonatal-seizures application asufficient amount of motion is needed to identify clinicalmanifestations like epileptic attacks In the application aboutbreathing disorders respiratory and abnormal movementsare needed for accurate diagnosis In the application aboutcongestive heart failures the motion is associated with thephysical activities of the patient Finally the amount ofmotion is associated with the medical staff activity ratherthan the patient for the application example related to theimprovement of medical practices in operating rooms
5 Conclusions
The proposed data hiding technique was demonstrated tobe suitable for the medicine healthcare context Our tech-nique successfully embedded samples of six EEG signalsinto encoded video sequences with high medium and lowmotion Our technique also extracted the hidden samplesfrom the encoded video sequences without loss of informa-tion The implementation of our technique required simplertasks compared to other existing synchronization techniques(1) less number of encoders and decoders (2) no timestampsneeded (3) no software needed for synchronization of videoand signal streams and (4) higher data capacity compared
to other data hiding techniques especially for high motionsequences
The experimental results demonstrated minimum degra-dation in video quality and data savings in terms of storagetransmission The experimental results for high and mediummotion video test sequences ranged from minus0007 dB to0011 dB in PSNR luma difference from 04459 to 13446in the bit-rate difference and from minus26196 to minus01273in embedded cost The changes in PSNR
119884difference and
bit-rate resulted in both no impacts in video-waveform syn-chronization and minimum distortions in video quality Forstorage and transmission purposes the embedded cost forhigh andmediummotion video sequences represent savingsFor low motion video sequences the experimental resultsranged from ndash0003 dB to minus0002 dB in PSNR
119884difference
from 333143 to 36575 in the bit-rate difference and from158725 to 193017 in embedded cost The changes in bit-rate were higher compared to the high and medium videosequences
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported by Latin American and CaribbeanCollaborative ICT Research Federation (LACCIR) through aresearch grant (R1209LAC001)
10 BioMed Research International
References
[1] S Bhattacharyya A Roy D P Dogra et al ldquoSummarizationof neonatal video EEG for seizure and artifact detectionrdquoin Proceedings of the 3rd National Conference on ComputerVision Pattern Recognition Image Processing and Graphics(NCVPRIPG rsquo11) pp 134ndash137 Hubli India December 2011
[2] J Zhang Y Li and Y Wei ldquoUsing timestamp to realize audio-video synchronization in real-time streaming media transmis-sionrdquo in Proceedings of the International Conference on AudioLanguage and Image Processing (ICALIP 08) pp 1073ndash1076Shanghai China July 2008
[3] S K Kapotas E E Varsaki and A N Skodras ldquoData hidingin H 264 encoded video sequencesrdquo in Proceedings of the 9thIEEE Workshop on Multimedia Signal Processing (MMSP rsquo07)pp 373ndash376 October 2007
[4] X Li H Chen D Wang and X Qi ldquoAudio-video synchronouscoding based on mode selection in H264rdquo in Proceedings of the4th International Congress on Image and Signal Processing (CISPrsquo11) vol 1 pp 113ndash117 October 2011
[5] B Li and M-Q Shi ldquoAudio-video synchronization codingapproach based on H264AVCrdquo IEICE Electronics Express vol6 no 22 pp 1556ndash1561 2009
[6] H-M Chao C-M Hsu and S-G Miaou ldquoA data-hiding tech-nique with authentication integration and confidentiality forelectronic patient recordsrdquo IEEE Transactions on InformationTechnology in Biomedicine vol 6 no 1 pp 46ndash53 2002
[7] C-H Liu and O T-C Chen ldquoData hiding in inter and intraprediction modes of H264AVCrdquo in Proceedings of the IEEEInternational Symposium on Circuits and Systems (ISCAS rsquo08)pp 3025ndash3028 May 2008
[8] K Liao D Ye S Lian Z Guo and J Wang ldquoLightweightinformation hiding inH264AVC video streamrdquo in Proceedingsof the International Conference on Multimedia InformationNetworking and Security vol 1 pp 578ndash582 November 2009
[9] H Zhu R Wang and D Xu ldquoInformation hiding algorithmfor H264 based on the motion estimation of quarter-pixelrdquoin Proceedings of the 2nd International Conference on FutureComputer and Communication (ICFCC rsquo10) vol 1 pp V1-423ndashV1-427 IEEE Wuhan China May 2010
[10] Y Li H-X Chen and Y Zhao ldquoA new method of data hidingbased on H264 encoded video sequencesrdquo in Proceedings of theIEEE 10th International Conference on Signal Processing (ICSPrsquo10) pp 1833ndash1836 October 2010
[11] P Wang Z Zheng and L Li ldquoA video watermarking schemebased on motion vectors and mode selectionrdquo in Proceedings ofthe International Conference on Computer Science and SoftwareEngineering (CSSE 08) vol 5 pp 233ndash237 Wuhan ChinaDecember 2008
[12] K Banno and M H Kryger ldquoUse of polysomnography withsynchronized digital video recording to diagnose pediatric sleepbreathing disordersrdquoCanadianMedical Association Journal vol173 no 1 pp 28ndash30 2005
[13] H J G van den Berg-Emons J B J Bussmann H M MBalk and H J Stam ldquoValidity of ambulatory accelerometry toquantify physical activity in heart failurerdquo Scandinavian Journalof Rehabilitation Medicine vol 32 no 4 pp 187ndash192 2000
[14] M B Weinger D C Gonzales J Slagle and M Syeed ldquoVideocapture of clinical care to enhance patient safetyrdquo Quality andSafety in Health Care vol 13 no 2 pp 136ndash144 2004
Figure 1 A comparison between (a) conventional synchronization process and (b) our proposed approach
multiplexing This technique maintains the correlationamong the media streams during the transmission processHowever multiplexing-based synchronization usually resultsin a loss of agility and integrality
Data hiding is another alternative for video-waveformsynchronization Data hiding has the goal of embeddinginformation into encoded video sequences with a minimumamount of perceivable degradation [3] The embedded infor-mation can be text pictures or physiological-signal samplesData hiding synchronizes the video and signals after hidingeach physiological sample at its corresponding video framein time
This paper introduces an improved data hiding techniquewith a larger data-hiding capacity in the context of medicalhealthcareOur data hiding technique synchronously embedsphysiological-signal samples intoH264AVC-encoded videosequencesThe implementation of data hiding is simpler thanother synchronization techniquesThe data-hiding techniquerequires only one encoder to process video and signalrsquossamples Data hiding also offers the advantage of a uniquecommunication channel for video-audio transmission andrequires no complex tasks related to timestamps synchro-nization marks or multiplexing [4 5] Figure 1 illustrates acomparison between the commonly used synchronizationprocesses and our synchronization approach based on datahiding Other important advantage is that our techniqueis strongly related to trends in secure handling of medicaldata during signal transmission Data hiding makes possible
the secure transmission of patientrsquos information over theinternet Important features for secure transmission of per-sonal information are authentication integrity and confiden-tiality [6]
2 Background
21 Data Hiding Techniques Existing data hiding techniquesare able to hide information in the video frames duringthe video encoding process In [3] authors proposed a datahiding scheme based on the macroblockrsquos size needed bythe H264AVC interprediction process The scheme is ableto hide two bits per macroblock and requires the followingpartitions types 16times 16 16times 8 8times 16 and 8times 8 The schemeloses no hidden data and may result in bit-rate increments
In [7] the data hiding scheme relies on constrainsassociated with the H264AVC interintraprediction modesIn the interprediction mode the scheme hides 0 bits at theinterprediction mode using the block sizes 16 times 8 8 times 168 times 4 and 4 times 8The scheme also hides a 1 bit using the blocksizes 16 times 16 8 times 8 and 4 times 4 In the intraprediction modethe scheme hides 0 bits using the block sizes 16times16 and 4times4Hiding a 1-bit value requires the 8 times 8 block size The schemehas minimum impact on the video quality and controls thedistortion degradation by hiding no data in 4 times 4 blocks
In [8] the data hiding scheme embeds information dur-ing context-based adaptive variable length coding (CAVLC)The scheme hides one bit using the trailing ones parity of
BioMed Research International 3
Table 1 Qualitative metrics and statistics of the data hiding (DH) schemes reviewed
Reference DH schemes DH capacity Max PSNR119884distortion Max bit-rate distortion
[3] Forcing block-type partitions 2 bits per MB (interframes) minus14 dB (40 kbps) 7000 bps[7] Grouping block-type partitions 1 bit per MB (interintraframes) minus009 dB 465[8] Parity of trailing ones CAVLC 1 bit per MB (intraframes) minus257 dB 0039[9] Quarter-pixel search positions 1ndash4 bits per MB (interframes) minus003 db 131[10] Last nonzero coefficient parity 1 bit per MB (interframes) Approx minus001 dB 1200 bps[11] Motion vectors and mode sel 1 bit per frame (interframes) 003 dB 098
the CAVLC code-word The CAVLC process results in mod-erate visual degradation and maintains the overall size of thevideo stream
In [9] the proposed technique exploited quarter-pixelmotion estimation process to hide dataThe schemehides onebit by modulating the best search points of a subblock Therate-distortion cost is introduced to reduce both the impacton the video quality and the increment in the bit-rate aftersearch point adjustments The hiding capacity is dependenton the content of the video sequence
In [10] the authors proposed a data hiding scheme basedon an adaptive method The method hides one bit using thelast nonzero coefficient parity after quantization of a 4 times 4luma block The scheme relies on an adaptive rather thana fixed point for data embedding The scheme results ina proportionally direct behavior between the bit-rate andcapacity size and between the bit-rate difference and theamount of embedded data
In [11] the proposed data hiding scheme relies onmotionvectors and mode selection The scheme only hides one bitper frame This scheme embeds data using the macroblocksearch regions with a left area restriction for a 1 bit and a rightarea restriction for 0 bits
In [4 5] the authors demonstrated that existing datahiding schemes in [8 10] successfully embedded audio intoencoded video sequences with minor impact on video imagequality and bit-rate
The desirable features of a data hiding technique suitablefor embedding physiological signals into encoded videosequences are a large data hiding capacity a low impact invideo quality and aminor effect in bit-rate of the video Table 1compares the data hiding schemes previously reviewed con-sidering three metrics hiding data capacity the maximumPSNR
119884(objective video quality) and bit-rate distortion
These schemes offer relatively low values for the threemetricsTherefore the main limitation resides in the data hidingcapacity In [9] the quarter-pixel motion estimation schemeoffers the highest data hiding capacity using an 8 times 8partitionThis scheme also offers very low PSNR
119884distortion and less
than 132 of bit-rate distortion Our proposed techniqueextends the quarter-pixelmotion estimation scheme to satisfythe data hiding capacity needs and to ensure the low PSNR
119884
and bit-rate distortions
22 Motion Estimation in H264AVC Motion estimation(ME) is an important element in the H264AVC interpre-diction process For a given frame the ME goal is to find
Table 2 Examples of medical applications
Reference Medical application Phys signals[1] Silent neonatal seizures EEG
[14] Patient safety in anesthesiaoperating rooms Vital signs
the best predictions for both levels macroblock (MB) selec-tion and motion vector (MV) estimation A MB is an arrayof 16 times 16 pixels The MB selection process assumes thepartitioning illustrated in Figure 2 Each partition containsa MV value Equation (1) shows how to select the bestblock partition by calculating the Lagrangian rate distortion(119869mode) optimization In this equation120582mode is the Lagrangianmultiplier SSD is the sum of the squared difference betweenthe original and the reconstructed block and119877 is the numberof bits of MB parameters such as quantization parameterheader motion vectors and residue coefficients
119869mode = SSD + (120582mode) (119877) (1)
The motion estimation process computes motion vectorsfor each macroblock partition found in each video frameAt a given frame the ME process searches for the new MBposition of each MB located in the reference frame The MEprocess calculates motion vectors based on these new MBpositions and encodes these vectors in the encoded framesFigure 2 illustrates the three ME stages to compute a MVThe first ME stage identifies the best MB position at theinteger-pixel mesh The second ME stage identifies the bestMB position at the half-pixel mesh based on the best integer-pixel position The third ME stage identifies the best MBposition at the quarter-pixelmesh based on the best half-pixelposition The selected position becomes the final MV value
23 Application Examples of Video-Based Medical Care Thissection presents additional examples of medical applica-tions related to the simultaneous observation of video andphysiological-signal waveforms Table 2 shows the names ofthe applications and the specific physiological signals neededfor simultaneous correlation with the video In [1] physicianstake advantage of synchronized EEG recordings with video to
4 BioMed Research International
Macroblock partitions Motion estimation (ME)
16 times 16
8 times 16
16 times 8
8 times 8
8 times 8 8 times 4
4 times 44 times 8
Stage 1Stage 2
Stage 3Selected
Submacroblock partitions
Figure 2 Macroblock partitions and search points in motion estimation
correlate clinical manifestations such as lip smacking fixingof eyeballs and cyclic leg movements
In [12] physicians take advantage of synchronized digitalvideo recordings to identify nocturnal breathing anomaliesusually undetected by standard polysomnography Successfulidentification of these anomalies requires the correlationamong EEG recordings oxygen saturation (SpO
2) endtidal
CO2level in-video leg movement and in-video rapid eye
movement (REM)In [13] the purpose is to assess the validity of a new
physical-activity monitor in the context of congestive heartfailure This monitor utilizes body-fixed accelerometers todistinguish among activities such as body postures sit-ting standing normal walking stairs walking cycling andwheelchair driving These in-video activities are correlatedwith the accelerometer waveforms to assess the correctoperation of the activity monitor
In [14] clinical investigators perform rigorous studies toenhance patient safety in operating rooms The investigatorsfirst elaborate a permanent feedback record containing the in-video health delivery process vital signs and other signalsThen the investigators reproduce this permanent recordto observe and to assess the health delivery process Thesynchronization of the video and the physiological-signalwaveforms is fundamental for the identification of factorsresulting in adverse events
3 The Proposed Data Hiding Technique
Our proposed technique hides streams of data samplesinto encoded video sequences The implementation of ourtechnique is a set of software routines added to the original
0 1 14 15 16 17 30 31
Time
EEG
Videoframes
middot middot middot
middot middot middot
middot middot middot
middot middot middot
fr = 30 framess
rEEG = 256 sampless
Figure 3 Synchronization process
H264AVC codec Figure 3 illustrates the synchronizationbetween the video and the physiological-signal waveformsof EEG samples Our technique synchronizes video andEEG signals by hiding samples of these signals at theircorresponding frame in time
31 Encoding Process in Our Data Hiding Technique Ourtechnique modifies the H264AVC interprediction processto hide the samples of the physiological signals Our tech-nique relies upon the quarter-pixel motion estimation pro-cess part of the H264AVC encoder to hide these samplesinto encoded video frames Our technique divides the searchpoints into four groups to improve the data hiding capacity ofthe original technique [9] Equation (2) presents the proposedpartitioning of search points and the binary assignment foreach partition The expression in (3) indicates how to select
BioMed Research International 5
Video
Transform PS sample
Mapping ME
MEno
sample
Motioncompensation
Current video frame
Video
Previouslycoded frames
Yes
No
force
frames
frames
14 pixel SP
Intra iquestallsamples
Intracoding Intercoding
intercoding
Figure 4 The block diagram of our proposed data hiding technique based on H264AVC encoder
Algorithm Embed a sampleInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(6) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(7) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(8) case 11 Quarter-Pixel Search Position = 0(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 1 Algorithm of our data hiding encoding technique
the search point in each group with the minimum distortioncost In (3) 119869119894 is the Lagrangian rate distortion parameter
Our proposed technique illustrated in Figure 4 repeatsthe gray blocks until no samples are available The blocksin gray are the additional routines needed to implementour proposed data hiding algorithm The alternate pathexecutes the original H264AVCmotion estimation processAlgorithm 1 presents details of our data hiding algorithmfor encoding Our technique hides the signal samples in themotion vectors of each block partition located at a frameEach motion vectors hides two bits of a signal sample Ourtechnique hides no samples in the block types I4MB andI16MB due to their association with intraframe prediction
Our technique is also unable to hide samples into PSKIPblocks due to the lack of motion vectors
Our technique also incorporates an approach to over-come the data-hiding capacity limitation found in low-motion video sequences PSKIP blocks are the most com-mon block partition found in encoded low-motion videosequences A large number of PSKIP blocks limit the data hid-ing capacity of our technique due to few number of motionvectors found in the low-motion video sequence Thereforeour proposed technique forces the H264AVC encoder toreplace the PSKIP block by a P16 times 16 block partitionThis PSKIP replacement adds a motion vector to the datahiding capacity of our techniqueThePSKIP replacement alsocontributes to maintaining the synchronization between theencoded video and the physiological signals However thisreplacement may also result in an increment in the bit-rate ofthe sequence Algorithm 2 presents details of the algorithmfor low-motion sequences
32 Decoding Process of Our Data Hiding Technique Ourdata hiding technique extracts the samples of the EEG signalat the same time conventional video H264AVC decoding
6 BioMed Research International
Algorithm Embed a sample in low-motion sequencesInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) if PSKIP partition do(4) curblock = P16 times 16 partition type(5) switchMapping do(6) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(7) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(8) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(9) case 11 Quarter-Pixel Search Position = 0(10) end switch(11) end ifrarr until complete all partitions of the current macroblock(12) end forrarr until complete all samples
Algorithm 2 Algorithm of our data hiding encoding technique for low motion sequences
Referencevideo
frames
Read MV ofMB partitions
Extractbit pairs
Physiologicalsignaloutput
Video
Encoded videosequence
No
YesConcatenated
value
iquestallsamples
Figure 5 Sample extraction during the decoding process
process takes placeThis is an important feature at the time todo synchronized playback of both video and physiological-signal waveforms The sample extraction algorithm in ourtechnique catches and gathers sample data This extractionalgorithm illustrated in Figure 5 is repeated as many timesas needed to extract all the samples embedded in the videosequence To do this our data hiding technique interacts withthe H264AVC decoding process
The routine reads the motion vector (MV) of every mac-roblock partition and inputs the MV
119909and MV
119910components
into (4) to identify a binary combinationThe |MV|2 termsoutput the quarter-pixel search point values The routineobtains the bit-pair samples after identifying the 1198661 to 1198664values mapped into (2)
type isin
1198661 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
1198662 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198663 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198664 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
(4)
Algorithm 3 presents details of our proposed decodingalgorithm This decoding process extracts the physiologicalsamples from the encoded video sequences
4 Results
Theexperimental setup illustrated in Figure 6 included a PCan EEG database and a set of seven video test sequencesThe EEG samples were extracted from the CHB-MIT ScalpEEGDatabaseThis database is located at the PhysioBankdig-ital recordings (httpwwwphysionetorg)The experimentalsetup included 6 signal electrodes at 256 samples per secondand 12-bit sample resolution [1] The EEG samples generatedin one second were embedded into the first 30 frames of eachtest video sequence to establish synchronization
The implementation of encoding and decoding processesneeded the modification of the JM reference software version161 Table 3 shows the JM configuration A program wasdeveloped to provide and convert samples from the databaseto the encoder
The video test sequences had a CIF (352times288) resolutionand a 420 YUV format The name of the test sequences are
BioMed Research International 7
Algorithm Extract a sampleInput Embedded sample in a synchronous frameResult Data of the sample in bit-pairs according to (2)(1) for ExtractSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case (|MV
119909|2 = 1) amp (|MV
119910|2 = 0) Bit-pair = 00
(6) case (|MV119909|2 = 0) amp (|MV
119910|2 = a) Bit-pair = 01
(7) case (|MV119909|2 = 1) amp (|MV
119910|2 = 1) Bit-pair = 10
(8) case (|MV119909|2 = 0) amp (|MV
119910|2 = 0) Bit-pair = 11
(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 3 Algorithm of our data hiding decoding technique
Experimental setup
Video testsequences
Neonatal
EEG
samples
Storage
Transmission
ModifiedJM reference
H264AVC
09998
09989
09985
09973
09944
09913
0 1 2 3
Time (s)
Figure 6 Experimental setup for our synchronization data hiding scheme
Table 3 Configuration parameters for JM reference software
Parameter ResolutionProfile BaselineFrames 30Motion estimation algorithm Full searchRD optimization and rate control Disabled8 times 8 subblocks DisabledNumber of reference frames 1Quantization parameter (QP) 28
akiyo bridge-far carphone football foreman mobile andneonatal Neonatal is not considered a standard video testsequence Neonatal was introduced to match the context ofthe application example related to EEG seizures on neonatesThe selected coding structure of the bit stream is ldquoIPPP rdquo tohave an intraframe encoded in the first frame and interframesencoded in the remaining frames
Metrics to evaluate the effectiveness of our proposed tech-nique are video objective quality bit-rate difference embed-ding cost and perceptual quality of the image The peak
signal-to-noise ratio (PSNR) illustrated in (5) is an objectivequality metric to report video image degradation119872 and 119873are the height and the width of the video frame respectively119868119894119895 and 1198681198941198951015840 represent the original pixel and the processedpixel respectively [5] The PSNR
119884diff metric illustrated in(6) indicates how the luma (119884) samples impact the videoquality after embedding the physiological samples PSNR1015840
119884
represents the impact of luma samples generated by embed-ding the samples and PSNR
119884represents the impact of luma
samples generated with the original H264AVC encoder
PSNR = 10 logmax (1198681198941198952)
(1119872)sum
119872
119894sum
119873
119895[119868119894119895 minus 119868119894119895
1015840]
2 (5)
PSNR119884diff = PSNR
1015840
119884minus PSNR
119884
(6)
Equation (7) shows how to calculate the change of bit-rate (BRI) 119877 is the original bit-rate and 1198771015840 is the embeddingsamples bit-rate Equation (8) shows how to estimate theembedded cost Oe Ov is the data volume generated by theoriginal video coder DHe is the data volume generated byembedding the samples and EEGe is the embedded datavolume [7] The EEGe term refers to the data coming from
8 BioMed Research International
Table 4 Experimental results for seven video test sequences
Figure 7 PSNR luma difference values for neonatal video sequence
Football Mobile Foreman Carphone Neonatal Akiyo Bridge-far
Video test sequences
High motionMedium motionLow motion
00005
0010015
0020025
0030035
minus001minus0005PS
NR Y
diffe
renc
e (dB
)
Figure 8 PSNR luma difference for seven video test sequences
the EEG signals The perceptual quality provides an estima-tion of subjective quality of the image obtained by visualinspection
BRI = 1198771015840minus 119877
119877
times 100 (7)
Oe = DHe minusOv minus EEGeOv + EEGe
times 100 (8)
Figure 7 compares the PSNR119884difference between the
original and the embedding data encoding processes Thegraph shows a very small difference between PSNR
119884values
for the first 30 frames of the neonatal video sequence Thelargest increment in quality was 0187 dB This incrementappeared in the 29th frame The largest decrement in qualitywas minus0039 dB This decrement appeared in the 8th frame
Figure 8 demonstrates that the embedding process hasminor effect on the video objective quality of seven testsequences The 119910-axis represents the PSNR luma differencevalues expressed in decibels The positive values above aver-age 0011 dB and 0029 dB represent a very small improve-ment in video qualityThe negative values from minus0002 dB to
05
10152025303540
Bit-r
ate d
iffer
ence
()
Carphone Neonatal Akiyo Bridge-farFootball Mobile Foreman
Video test sequences
High motionMedium motionLow motion
Figure 9 Change of bit-rate for seven video test sequences
minus0007 represent a small degradation in video quality Thesedegradations and improvements are due to the lack of bit-rateconstraint
Embedding the EEG data into the video sequencesproduced increments of bit-rate for the seven video testsequences as illustrated in Figure 9 Our experimental setupincluded high motion medium motion and low motionvideo test sequences The amount motion of the neonatalvideo sequence was considered between medium and lowThe graph presents the neonatal sequence with gray colorto indicate that it is not a standard video test sequence Forthe high and medium motion sequences there are small bit-rate increments The bit-rate increments ranged from 045to 134 For the neonatal and low motion sequences thePSKIP block replacement occurred and resulted in bigger bit-rate incrementsThe bit-rate increments ranged from 56 to3658 However the changes in the bit-rate had no effect onthe video-waveform synchronization
Table 4 shows the results for the seven test sequences interms of embedded capacity modified macroblocks mod-ified motion vectors PSNR
119884difference bit-rate and the
embedded cost For the high medium motion and neonatalsequences the embedded cost ranged from minus01273 tominus26196 representing savings in data volume For the lowmotion sequences the embedded cost had an incrementin 158726 and 193017 These results indicate that ourdata hiding technique offers both an adequate efficiency forvideo-signal transmission and savings in the storage of highmedium and neonatal sequences
The inspection of subject quality indicated minimumvisual artifacts or distortion between original and data-hiding
BioMed Research International 9
(a) (b)
(c)
Figure 10 10th frame of neonatal video sequence (a) Original image (b) image using video original coding and (c) image with EEG data
images Figure 10 illustrates the perceptual quality of thevideo frames for the 10th frame of the neonatal sequenceThus data hiding generates no significant difference inquality from a human eye perception
Our technique will offer an adequate performance in thecontext of the application examples presented in Section 23The video sequences of these applications exhibit sufficientamount of motion In the neonatal-seizures application asufficient amount of motion is needed to identify clinicalmanifestations like epileptic attacks In the application aboutbreathing disorders respiratory and abnormal movementsare needed for accurate diagnosis In the application aboutcongestive heart failures the motion is associated with thephysical activities of the patient Finally the amount ofmotion is associated with the medical staff activity ratherthan the patient for the application example related to theimprovement of medical practices in operating rooms
5 Conclusions
The proposed data hiding technique was demonstrated tobe suitable for the medicine healthcare context Our tech-nique successfully embedded samples of six EEG signalsinto encoded video sequences with high medium and lowmotion Our technique also extracted the hidden samplesfrom the encoded video sequences without loss of informa-tion The implementation of our technique required simplertasks compared to other existing synchronization techniques(1) less number of encoders and decoders (2) no timestampsneeded (3) no software needed for synchronization of videoand signal streams and (4) higher data capacity compared
to other data hiding techniques especially for high motionsequences
The experimental results demonstrated minimum degra-dation in video quality and data savings in terms of storagetransmission The experimental results for high and mediummotion video test sequences ranged from minus0007 dB to0011 dB in PSNR luma difference from 04459 to 13446in the bit-rate difference and from minus26196 to minus01273in embedded cost The changes in PSNR
119884difference and
bit-rate resulted in both no impacts in video-waveform syn-chronization and minimum distortions in video quality Forstorage and transmission purposes the embedded cost forhigh andmediummotion video sequences represent savingsFor low motion video sequences the experimental resultsranged from ndash0003 dB to minus0002 dB in PSNR
119884difference
from 333143 to 36575 in the bit-rate difference and from158725 to 193017 in embedded cost The changes in bit-rate were higher compared to the high and medium videosequences
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported by Latin American and CaribbeanCollaborative ICT Research Federation (LACCIR) through aresearch grant (R1209LAC001)
10 BioMed Research International
References
[1] S Bhattacharyya A Roy D P Dogra et al ldquoSummarizationof neonatal video EEG for seizure and artifact detectionrdquoin Proceedings of the 3rd National Conference on ComputerVision Pattern Recognition Image Processing and Graphics(NCVPRIPG rsquo11) pp 134ndash137 Hubli India December 2011
[2] J Zhang Y Li and Y Wei ldquoUsing timestamp to realize audio-video synchronization in real-time streaming media transmis-sionrdquo in Proceedings of the International Conference on AudioLanguage and Image Processing (ICALIP 08) pp 1073ndash1076Shanghai China July 2008
[3] S K Kapotas E E Varsaki and A N Skodras ldquoData hidingin H 264 encoded video sequencesrdquo in Proceedings of the 9thIEEE Workshop on Multimedia Signal Processing (MMSP rsquo07)pp 373ndash376 October 2007
[4] X Li H Chen D Wang and X Qi ldquoAudio-video synchronouscoding based on mode selection in H264rdquo in Proceedings of the4th International Congress on Image and Signal Processing (CISPrsquo11) vol 1 pp 113ndash117 October 2011
[5] B Li and M-Q Shi ldquoAudio-video synchronization codingapproach based on H264AVCrdquo IEICE Electronics Express vol6 no 22 pp 1556ndash1561 2009
[6] H-M Chao C-M Hsu and S-G Miaou ldquoA data-hiding tech-nique with authentication integration and confidentiality forelectronic patient recordsrdquo IEEE Transactions on InformationTechnology in Biomedicine vol 6 no 1 pp 46ndash53 2002
[7] C-H Liu and O T-C Chen ldquoData hiding in inter and intraprediction modes of H264AVCrdquo in Proceedings of the IEEEInternational Symposium on Circuits and Systems (ISCAS rsquo08)pp 3025ndash3028 May 2008
[8] K Liao D Ye S Lian Z Guo and J Wang ldquoLightweightinformation hiding inH264AVC video streamrdquo in Proceedingsof the International Conference on Multimedia InformationNetworking and Security vol 1 pp 578ndash582 November 2009
[9] H Zhu R Wang and D Xu ldquoInformation hiding algorithmfor H264 based on the motion estimation of quarter-pixelrdquoin Proceedings of the 2nd International Conference on FutureComputer and Communication (ICFCC rsquo10) vol 1 pp V1-423ndashV1-427 IEEE Wuhan China May 2010
[10] Y Li H-X Chen and Y Zhao ldquoA new method of data hidingbased on H264 encoded video sequencesrdquo in Proceedings of theIEEE 10th International Conference on Signal Processing (ICSPrsquo10) pp 1833ndash1836 October 2010
[11] P Wang Z Zheng and L Li ldquoA video watermarking schemebased on motion vectors and mode selectionrdquo in Proceedings ofthe International Conference on Computer Science and SoftwareEngineering (CSSE 08) vol 5 pp 233ndash237 Wuhan ChinaDecember 2008
[12] K Banno and M H Kryger ldquoUse of polysomnography withsynchronized digital video recording to diagnose pediatric sleepbreathing disordersrdquoCanadianMedical Association Journal vol173 no 1 pp 28ndash30 2005
[13] H J G van den Berg-Emons J B J Bussmann H M MBalk and H J Stam ldquoValidity of ambulatory accelerometry toquantify physical activity in heart failurerdquo Scandinavian Journalof Rehabilitation Medicine vol 32 no 4 pp 187ndash192 2000
[14] M B Weinger D C Gonzales J Slagle and M Syeed ldquoVideocapture of clinical care to enhance patient safetyrdquo Quality andSafety in Health Care vol 13 no 2 pp 136ndash144 2004
Table 1 Qualitative metrics and statistics of the data hiding (DH) schemes reviewed
Reference DH schemes DH capacity Max PSNR119884distortion Max bit-rate distortion
[3] Forcing block-type partitions 2 bits per MB (interframes) minus14 dB (40 kbps) 7000 bps[7] Grouping block-type partitions 1 bit per MB (interintraframes) minus009 dB 465[8] Parity of trailing ones CAVLC 1 bit per MB (intraframes) minus257 dB 0039[9] Quarter-pixel search positions 1ndash4 bits per MB (interframes) minus003 db 131[10] Last nonzero coefficient parity 1 bit per MB (interframes) Approx minus001 dB 1200 bps[11] Motion vectors and mode sel 1 bit per frame (interframes) 003 dB 098
the CAVLC code-word The CAVLC process results in mod-erate visual degradation and maintains the overall size of thevideo stream
In [9] the proposed technique exploited quarter-pixelmotion estimation process to hide dataThe schemehides onebit by modulating the best search points of a subblock Therate-distortion cost is introduced to reduce both the impacton the video quality and the increment in the bit-rate aftersearch point adjustments The hiding capacity is dependenton the content of the video sequence
In [10] the authors proposed a data hiding scheme basedon an adaptive method The method hides one bit using thelast nonzero coefficient parity after quantization of a 4 times 4luma block The scheme relies on an adaptive rather thana fixed point for data embedding The scheme results ina proportionally direct behavior between the bit-rate andcapacity size and between the bit-rate difference and theamount of embedded data
In [11] the proposed data hiding scheme relies onmotionvectors and mode selection The scheme only hides one bitper frame This scheme embeds data using the macroblocksearch regions with a left area restriction for a 1 bit and a rightarea restriction for 0 bits
In [4 5] the authors demonstrated that existing datahiding schemes in [8 10] successfully embedded audio intoencoded video sequences with minor impact on video imagequality and bit-rate
The desirable features of a data hiding technique suitablefor embedding physiological signals into encoded videosequences are a large data hiding capacity a low impact invideo quality and aminor effect in bit-rate of the video Table 1compares the data hiding schemes previously reviewed con-sidering three metrics hiding data capacity the maximumPSNR
119884(objective video quality) and bit-rate distortion
These schemes offer relatively low values for the threemetricsTherefore the main limitation resides in the data hidingcapacity In [9] the quarter-pixel motion estimation schemeoffers the highest data hiding capacity using an 8 times 8partitionThis scheme also offers very low PSNR
119884distortion and less
than 132 of bit-rate distortion Our proposed techniqueextends the quarter-pixelmotion estimation scheme to satisfythe data hiding capacity needs and to ensure the low PSNR
119884
and bit-rate distortions
22 Motion Estimation in H264AVC Motion estimation(ME) is an important element in the H264AVC interpre-diction process For a given frame the ME goal is to find
Table 2 Examples of medical applications
Reference Medical application Phys signals[1] Silent neonatal seizures EEG
[14] Patient safety in anesthesiaoperating rooms Vital signs
the best predictions for both levels macroblock (MB) selec-tion and motion vector (MV) estimation A MB is an arrayof 16 times 16 pixels The MB selection process assumes thepartitioning illustrated in Figure 2 Each partition containsa MV value Equation (1) shows how to select the bestblock partition by calculating the Lagrangian rate distortion(119869mode) optimization In this equation120582mode is the Lagrangianmultiplier SSD is the sum of the squared difference betweenthe original and the reconstructed block and119877 is the numberof bits of MB parameters such as quantization parameterheader motion vectors and residue coefficients
119869mode = SSD + (120582mode) (119877) (1)
The motion estimation process computes motion vectorsfor each macroblock partition found in each video frameAt a given frame the ME process searches for the new MBposition of each MB located in the reference frame The MEprocess calculates motion vectors based on these new MBpositions and encodes these vectors in the encoded framesFigure 2 illustrates the three ME stages to compute a MVThe first ME stage identifies the best MB position at theinteger-pixel mesh The second ME stage identifies the bestMB position at the half-pixel mesh based on the best integer-pixel position The third ME stage identifies the best MBposition at the quarter-pixelmesh based on the best half-pixelposition The selected position becomes the final MV value
23 Application Examples of Video-Based Medical Care Thissection presents additional examples of medical applica-tions related to the simultaneous observation of video andphysiological-signal waveforms Table 2 shows the names ofthe applications and the specific physiological signals neededfor simultaneous correlation with the video In [1] physicianstake advantage of synchronized EEG recordings with video to
4 BioMed Research International
Macroblock partitions Motion estimation (ME)
16 times 16
8 times 16
16 times 8
8 times 8
8 times 8 8 times 4
4 times 44 times 8
Stage 1Stage 2
Stage 3Selected
Submacroblock partitions
Figure 2 Macroblock partitions and search points in motion estimation
correlate clinical manifestations such as lip smacking fixingof eyeballs and cyclic leg movements
In [12] physicians take advantage of synchronized digitalvideo recordings to identify nocturnal breathing anomaliesusually undetected by standard polysomnography Successfulidentification of these anomalies requires the correlationamong EEG recordings oxygen saturation (SpO
2) endtidal
CO2level in-video leg movement and in-video rapid eye
movement (REM)In [13] the purpose is to assess the validity of a new
physical-activity monitor in the context of congestive heartfailure This monitor utilizes body-fixed accelerometers todistinguish among activities such as body postures sit-ting standing normal walking stairs walking cycling andwheelchair driving These in-video activities are correlatedwith the accelerometer waveforms to assess the correctoperation of the activity monitor
In [14] clinical investigators perform rigorous studies toenhance patient safety in operating rooms The investigatorsfirst elaborate a permanent feedback record containing the in-video health delivery process vital signs and other signalsThen the investigators reproduce this permanent recordto observe and to assess the health delivery process Thesynchronization of the video and the physiological-signalwaveforms is fundamental for the identification of factorsresulting in adverse events
3 The Proposed Data Hiding Technique
Our proposed technique hides streams of data samplesinto encoded video sequences The implementation of ourtechnique is a set of software routines added to the original
0 1 14 15 16 17 30 31
Time
EEG
Videoframes
middot middot middot
middot middot middot
middot middot middot
middot middot middot
fr = 30 framess
rEEG = 256 sampless
Figure 3 Synchronization process
H264AVC codec Figure 3 illustrates the synchronizationbetween the video and the physiological-signal waveformsof EEG samples Our technique synchronizes video andEEG signals by hiding samples of these signals at theircorresponding frame in time
31 Encoding Process in Our Data Hiding Technique Ourtechnique modifies the H264AVC interprediction processto hide the samples of the physiological signals Our tech-nique relies upon the quarter-pixel motion estimation pro-cess part of the H264AVC encoder to hide these samplesinto encoded video frames Our technique divides the searchpoints into four groups to improve the data hiding capacity ofthe original technique [9] Equation (2) presents the proposedpartitioning of search points and the binary assignment foreach partition The expression in (3) indicates how to select
BioMed Research International 5
Video
Transform PS sample
Mapping ME
MEno
sample
Motioncompensation
Current video frame
Video
Previouslycoded frames
Yes
No
force
frames
frames
14 pixel SP
Intra iquestallsamples
Intracoding Intercoding
intercoding
Figure 4 The block diagram of our proposed data hiding technique based on H264AVC encoder
Algorithm Embed a sampleInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(6) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(7) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(8) case 11 Quarter-Pixel Search Position = 0(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 1 Algorithm of our data hiding encoding technique
the search point in each group with the minimum distortioncost In (3) 119869119894 is the Lagrangian rate distortion parameter
Our proposed technique illustrated in Figure 4 repeatsthe gray blocks until no samples are available The blocksin gray are the additional routines needed to implementour proposed data hiding algorithm The alternate pathexecutes the original H264AVCmotion estimation processAlgorithm 1 presents details of our data hiding algorithmfor encoding Our technique hides the signal samples in themotion vectors of each block partition located at a frameEach motion vectors hides two bits of a signal sample Ourtechnique hides no samples in the block types I4MB andI16MB due to their association with intraframe prediction
Our technique is also unable to hide samples into PSKIPblocks due to the lack of motion vectors
Our technique also incorporates an approach to over-come the data-hiding capacity limitation found in low-motion video sequences PSKIP blocks are the most com-mon block partition found in encoded low-motion videosequences A large number of PSKIP blocks limit the data hid-ing capacity of our technique due to few number of motionvectors found in the low-motion video sequence Thereforeour proposed technique forces the H264AVC encoder toreplace the PSKIP block by a P16 times 16 block partitionThis PSKIP replacement adds a motion vector to the datahiding capacity of our techniqueThePSKIP replacement alsocontributes to maintaining the synchronization between theencoded video and the physiological signals However thisreplacement may also result in an increment in the bit-rate ofthe sequence Algorithm 2 presents details of the algorithmfor low-motion sequences
32 Decoding Process of Our Data Hiding Technique Ourdata hiding technique extracts the samples of the EEG signalat the same time conventional video H264AVC decoding
6 BioMed Research International
Algorithm Embed a sample in low-motion sequencesInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) if PSKIP partition do(4) curblock = P16 times 16 partition type(5) switchMapping do(6) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(7) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(8) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(9) case 11 Quarter-Pixel Search Position = 0(10) end switch(11) end ifrarr until complete all partitions of the current macroblock(12) end forrarr until complete all samples
Algorithm 2 Algorithm of our data hiding encoding technique for low motion sequences
Referencevideo
frames
Read MV ofMB partitions
Extractbit pairs
Physiologicalsignaloutput
Video
Encoded videosequence
No
YesConcatenated
value
iquestallsamples
Figure 5 Sample extraction during the decoding process
process takes placeThis is an important feature at the time todo synchronized playback of both video and physiological-signal waveforms The sample extraction algorithm in ourtechnique catches and gathers sample data This extractionalgorithm illustrated in Figure 5 is repeated as many timesas needed to extract all the samples embedded in the videosequence To do this our data hiding technique interacts withthe H264AVC decoding process
The routine reads the motion vector (MV) of every mac-roblock partition and inputs the MV
119909and MV
119910components
into (4) to identify a binary combinationThe |MV|2 termsoutput the quarter-pixel search point values The routineobtains the bit-pair samples after identifying the 1198661 to 1198664values mapped into (2)
type isin
1198661 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
1198662 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198663 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198664 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
(4)
Algorithm 3 presents details of our proposed decodingalgorithm This decoding process extracts the physiologicalsamples from the encoded video sequences
4 Results
Theexperimental setup illustrated in Figure 6 included a PCan EEG database and a set of seven video test sequencesThe EEG samples were extracted from the CHB-MIT ScalpEEGDatabaseThis database is located at the PhysioBankdig-ital recordings (httpwwwphysionetorg)The experimentalsetup included 6 signal electrodes at 256 samples per secondand 12-bit sample resolution [1] The EEG samples generatedin one second were embedded into the first 30 frames of eachtest video sequence to establish synchronization
The implementation of encoding and decoding processesneeded the modification of the JM reference software version161 Table 3 shows the JM configuration A program wasdeveloped to provide and convert samples from the databaseto the encoder
The video test sequences had a CIF (352times288) resolutionand a 420 YUV format The name of the test sequences are
BioMed Research International 7
Algorithm Extract a sampleInput Embedded sample in a synchronous frameResult Data of the sample in bit-pairs according to (2)(1) for ExtractSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case (|MV
119909|2 = 1) amp (|MV
119910|2 = 0) Bit-pair = 00
(6) case (|MV119909|2 = 0) amp (|MV
119910|2 = a) Bit-pair = 01
(7) case (|MV119909|2 = 1) amp (|MV
119910|2 = 1) Bit-pair = 10
(8) case (|MV119909|2 = 0) amp (|MV
119910|2 = 0) Bit-pair = 11
(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 3 Algorithm of our data hiding decoding technique
Experimental setup
Video testsequences
Neonatal
EEG
samples
Storage
Transmission
ModifiedJM reference
H264AVC
09998
09989
09985
09973
09944
09913
0 1 2 3
Time (s)
Figure 6 Experimental setup for our synchronization data hiding scheme
Table 3 Configuration parameters for JM reference software
Parameter ResolutionProfile BaselineFrames 30Motion estimation algorithm Full searchRD optimization and rate control Disabled8 times 8 subblocks DisabledNumber of reference frames 1Quantization parameter (QP) 28
akiyo bridge-far carphone football foreman mobile andneonatal Neonatal is not considered a standard video testsequence Neonatal was introduced to match the context ofthe application example related to EEG seizures on neonatesThe selected coding structure of the bit stream is ldquoIPPP rdquo tohave an intraframe encoded in the first frame and interframesencoded in the remaining frames
Metrics to evaluate the effectiveness of our proposed tech-nique are video objective quality bit-rate difference embed-ding cost and perceptual quality of the image The peak
signal-to-noise ratio (PSNR) illustrated in (5) is an objectivequality metric to report video image degradation119872 and 119873are the height and the width of the video frame respectively119868119894119895 and 1198681198941198951015840 represent the original pixel and the processedpixel respectively [5] The PSNR
119884diff metric illustrated in(6) indicates how the luma (119884) samples impact the videoquality after embedding the physiological samples PSNR1015840
119884
represents the impact of luma samples generated by embed-ding the samples and PSNR
119884represents the impact of luma
samples generated with the original H264AVC encoder
PSNR = 10 logmax (1198681198941198952)
(1119872)sum
119872
119894sum
119873
119895[119868119894119895 minus 119868119894119895
1015840]
2 (5)
PSNR119884diff = PSNR
1015840
119884minus PSNR
119884
(6)
Equation (7) shows how to calculate the change of bit-rate (BRI) 119877 is the original bit-rate and 1198771015840 is the embeddingsamples bit-rate Equation (8) shows how to estimate theembedded cost Oe Ov is the data volume generated by theoriginal video coder DHe is the data volume generated byembedding the samples and EEGe is the embedded datavolume [7] The EEGe term refers to the data coming from
8 BioMed Research International
Table 4 Experimental results for seven video test sequences
Figure 7 PSNR luma difference values for neonatal video sequence
Football Mobile Foreman Carphone Neonatal Akiyo Bridge-far
Video test sequences
High motionMedium motionLow motion
00005
0010015
0020025
0030035
minus001minus0005PS
NR Y
diffe
renc
e (dB
)
Figure 8 PSNR luma difference for seven video test sequences
the EEG signals The perceptual quality provides an estima-tion of subjective quality of the image obtained by visualinspection
BRI = 1198771015840minus 119877
119877
times 100 (7)
Oe = DHe minusOv minus EEGeOv + EEGe
times 100 (8)
Figure 7 compares the PSNR119884difference between the
original and the embedding data encoding processes Thegraph shows a very small difference between PSNR
119884values
for the first 30 frames of the neonatal video sequence Thelargest increment in quality was 0187 dB This incrementappeared in the 29th frame The largest decrement in qualitywas minus0039 dB This decrement appeared in the 8th frame
Figure 8 demonstrates that the embedding process hasminor effect on the video objective quality of seven testsequences The 119910-axis represents the PSNR luma differencevalues expressed in decibels The positive values above aver-age 0011 dB and 0029 dB represent a very small improve-ment in video qualityThe negative values from minus0002 dB to
05
10152025303540
Bit-r
ate d
iffer
ence
()
Carphone Neonatal Akiyo Bridge-farFootball Mobile Foreman
Video test sequences
High motionMedium motionLow motion
Figure 9 Change of bit-rate for seven video test sequences
minus0007 represent a small degradation in video quality Thesedegradations and improvements are due to the lack of bit-rateconstraint
Embedding the EEG data into the video sequencesproduced increments of bit-rate for the seven video testsequences as illustrated in Figure 9 Our experimental setupincluded high motion medium motion and low motionvideo test sequences The amount motion of the neonatalvideo sequence was considered between medium and lowThe graph presents the neonatal sequence with gray colorto indicate that it is not a standard video test sequence Forthe high and medium motion sequences there are small bit-rate increments The bit-rate increments ranged from 045to 134 For the neonatal and low motion sequences thePSKIP block replacement occurred and resulted in bigger bit-rate incrementsThe bit-rate increments ranged from 56 to3658 However the changes in the bit-rate had no effect onthe video-waveform synchronization
Table 4 shows the results for the seven test sequences interms of embedded capacity modified macroblocks mod-ified motion vectors PSNR
119884difference bit-rate and the
embedded cost For the high medium motion and neonatalsequences the embedded cost ranged from minus01273 tominus26196 representing savings in data volume For the lowmotion sequences the embedded cost had an incrementin 158726 and 193017 These results indicate that ourdata hiding technique offers both an adequate efficiency forvideo-signal transmission and savings in the storage of highmedium and neonatal sequences
The inspection of subject quality indicated minimumvisual artifacts or distortion between original and data-hiding
BioMed Research International 9
(a) (b)
(c)
Figure 10 10th frame of neonatal video sequence (a) Original image (b) image using video original coding and (c) image with EEG data
images Figure 10 illustrates the perceptual quality of thevideo frames for the 10th frame of the neonatal sequenceThus data hiding generates no significant difference inquality from a human eye perception
Our technique will offer an adequate performance in thecontext of the application examples presented in Section 23The video sequences of these applications exhibit sufficientamount of motion In the neonatal-seizures application asufficient amount of motion is needed to identify clinicalmanifestations like epileptic attacks In the application aboutbreathing disorders respiratory and abnormal movementsare needed for accurate diagnosis In the application aboutcongestive heart failures the motion is associated with thephysical activities of the patient Finally the amount ofmotion is associated with the medical staff activity ratherthan the patient for the application example related to theimprovement of medical practices in operating rooms
5 Conclusions
The proposed data hiding technique was demonstrated tobe suitable for the medicine healthcare context Our tech-nique successfully embedded samples of six EEG signalsinto encoded video sequences with high medium and lowmotion Our technique also extracted the hidden samplesfrom the encoded video sequences without loss of informa-tion The implementation of our technique required simplertasks compared to other existing synchronization techniques(1) less number of encoders and decoders (2) no timestampsneeded (3) no software needed for synchronization of videoand signal streams and (4) higher data capacity compared
to other data hiding techniques especially for high motionsequences
The experimental results demonstrated minimum degra-dation in video quality and data savings in terms of storagetransmission The experimental results for high and mediummotion video test sequences ranged from minus0007 dB to0011 dB in PSNR luma difference from 04459 to 13446in the bit-rate difference and from minus26196 to minus01273in embedded cost The changes in PSNR
119884difference and
bit-rate resulted in both no impacts in video-waveform syn-chronization and minimum distortions in video quality Forstorage and transmission purposes the embedded cost forhigh andmediummotion video sequences represent savingsFor low motion video sequences the experimental resultsranged from ndash0003 dB to minus0002 dB in PSNR
119884difference
from 333143 to 36575 in the bit-rate difference and from158725 to 193017 in embedded cost The changes in bit-rate were higher compared to the high and medium videosequences
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported by Latin American and CaribbeanCollaborative ICT Research Federation (LACCIR) through aresearch grant (R1209LAC001)
10 BioMed Research International
References
[1] S Bhattacharyya A Roy D P Dogra et al ldquoSummarizationof neonatal video EEG for seizure and artifact detectionrdquoin Proceedings of the 3rd National Conference on ComputerVision Pattern Recognition Image Processing and Graphics(NCVPRIPG rsquo11) pp 134ndash137 Hubli India December 2011
[2] J Zhang Y Li and Y Wei ldquoUsing timestamp to realize audio-video synchronization in real-time streaming media transmis-sionrdquo in Proceedings of the International Conference on AudioLanguage and Image Processing (ICALIP 08) pp 1073ndash1076Shanghai China July 2008
[3] S K Kapotas E E Varsaki and A N Skodras ldquoData hidingin H 264 encoded video sequencesrdquo in Proceedings of the 9thIEEE Workshop on Multimedia Signal Processing (MMSP rsquo07)pp 373ndash376 October 2007
[4] X Li H Chen D Wang and X Qi ldquoAudio-video synchronouscoding based on mode selection in H264rdquo in Proceedings of the4th International Congress on Image and Signal Processing (CISPrsquo11) vol 1 pp 113ndash117 October 2011
[5] B Li and M-Q Shi ldquoAudio-video synchronization codingapproach based on H264AVCrdquo IEICE Electronics Express vol6 no 22 pp 1556ndash1561 2009
[6] H-M Chao C-M Hsu and S-G Miaou ldquoA data-hiding tech-nique with authentication integration and confidentiality forelectronic patient recordsrdquo IEEE Transactions on InformationTechnology in Biomedicine vol 6 no 1 pp 46ndash53 2002
[7] C-H Liu and O T-C Chen ldquoData hiding in inter and intraprediction modes of H264AVCrdquo in Proceedings of the IEEEInternational Symposium on Circuits and Systems (ISCAS rsquo08)pp 3025ndash3028 May 2008
[8] K Liao D Ye S Lian Z Guo and J Wang ldquoLightweightinformation hiding inH264AVC video streamrdquo in Proceedingsof the International Conference on Multimedia InformationNetworking and Security vol 1 pp 578ndash582 November 2009
[9] H Zhu R Wang and D Xu ldquoInformation hiding algorithmfor H264 based on the motion estimation of quarter-pixelrdquoin Proceedings of the 2nd International Conference on FutureComputer and Communication (ICFCC rsquo10) vol 1 pp V1-423ndashV1-427 IEEE Wuhan China May 2010
[10] Y Li H-X Chen and Y Zhao ldquoA new method of data hidingbased on H264 encoded video sequencesrdquo in Proceedings of theIEEE 10th International Conference on Signal Processing (ICSPrsquo10) pp 1833ndash1836 October 2010
[11] P Wang Z Zheng and L Li ldquoA video watermarking schemebased on motion vectors and mode selectionrdquo in Proceedings ofthe International Conference on Computer Science and SoftwareEngineering (CSSE 08) vol 5 pp 233ndash237 Wuhan ChinaDecember 2008
[12] K Banno and M H Kryger ldquoUse of polysomnography withsynchronized digital video recording to diagnose pediatric sleepbreathing disordersrdquoCanadianMedical Association Journal vol173 no 1 pp 28ndash30 2005
[13] H J G van den Berg-Emons J B J Bussmann H M MBalk and H J Stam ldquoValidity of ambulatory accelerometry toquantify physical activity in heart failurerdquo Scandinavian Journalof Rehabilitation Medicine vol 32 no 4 pp 187ndash192 2000
[14] M B Weinger D C Gonzales J Slagle and M Syeed ldquoVideocapture of clinical care to enhance patient safetyrdquo Quality andSafety in Health Care vol 13 no 2 pp 136ndash144 2004
Figure 2 Macroblock partitions and search points in motion estimation
correlate clinical manifestations such as lip smacking fixingof eyeballs and cyclic leg movements
In [12] physicians take advantage of synchronized digitalvideo recordings to identify nocturnal breathing anomaliesusually undetected by standard polysomnography Successfulidentification of these anomalies requires the correlationamong EEG recordings oxygen saturation (SpO
2) endtidal
CO2level in-video leg movement and in-video rapid eye
movement (REM)In [13] the purpose is to assess the validity of a new
physical-activity monitor in the context of congestive heartfailure This monitor utilizes body-fixed accelerometers todistinguish among activities such as body postures sit-ting standing normal walking stairs walking cycling andwheelchair driving These in-video activities are correlatedwith the accelerometer waveforms to assess the correctoperation of the activity monitor
In [14] clinical investigators perform rigorous studies toenhance patient safety in operating rooms The investigatorsfirst elaborate a permanent feedback record containing the in-video health delivery process vital signs and other signalsThen the investigators reproduce this permanent recordto observe and to assess the health delivery process Thesynchronization of the video and the physiological-signalwaveforms is fundamental for the identification of factorsresulting in adverse events
3 The Proposed Data Hiding Technique
Our proposed technique hides streams of data samplesinto encoded video sequences The implementation of ourtechnique is a set of software routines added to the original
0 1 14 15 16 17 30 31
Time
EEG
Videoframes
middot middot middot
middot middot middot
middot middot middot
middot middot middot
fr = 30 framess
rEEG = 256 sampless
Figure 3 Synchronization process
H264AVC codec Figure 3 illustrates the synchronizationbetween the video and the physiological-signal waveformsof EEG samples Our technique synchronizes video andEEG signals by hiding samples of these signals at theircorresponding frame in time
31 Encoding Process in Our Data Hiding Technique Ourtechnique modifies the H264AVC interprediction processto hide the samples of the physiological signals Our tech-nique relies upon the quarter-pixel motion estimation pro-cess part of the H264AVC encoder to hide these samplesinto encoded video frames Our technique divides the searchpoints into four groups to improve the data hiding capacity ofthe original technique [9] Equation (2) presents the proposedpartitioning of search points and the binary assignment foreach partition The expression in (3) indicates how to select
BioMed Research International 5
Video
Transform PS sample
Mapping ME
MEno
sample
Motioncompensation
Current video frame
Video
Previouslycoded frames
Yes
No
force
frames
frames
14 pixel SP
Intra iquestallsamples
Intracoding Intercoding
intercoding
Figure 4 The block diagram of our proposed data hiding technique based on H264AVC encoder
Algorithm Embed a sampleInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(6) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(7) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(8) case 11 Quarter-Pixel Search Position = 0(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 1 Algorithm of our data hiding encoding technique
the search point in each group with the minimum distortioncost In (3) 119869119894 is the Lagrangian rate distortion parameter
Our proposed technique illustrated in Figure 4 repeatsthe gray blocks until no samples are available The blocksin gray are the additional routines needed to implementour proposed data hiding algorithm The alternate pathexecutes the original H264AVCmotion estimation processAlgorithm 1 presents details of our data hiding algorithmfor encoding Our technique hides the signal samples in themotion vectors of each block partition located at a frameEach motion vectors hides two bits of a signal sample Ourtechnique hides no samples in the block types I4MB andI16MB due to their association with intraframe prediction
Our technique is also unable to hide samples into PSKIPblocks due to the lack of motion vectors
Our technique also incorporates an approach to over-come the data-hiding capacity limitation found in low-motion video sequences PSKIP blocks are the most com-mon block partition found in encoded low-motion videosequences A large number of PSKIP blocks limit the data hid-ing capacity of our technique due to few number of motionvectors found in the low-motion video sequence Thereforeour proposed technique forces the H264AVC encoder toreplace the PSKIP block by a P16 times 16 block partitionThis PSKIP replacement adds a motion vector to the datahiding capacity of our techniqueThePSKIP replacement alsocontributes to maintaining the synchronization between theencoded video and the physiological signals However thisreplacement may also result in an increment in the bit-rate ofthe sequence Algorithm 2 presents details of the algorithmfor low-motion sequences
32 Decoding Process of Our Data Hiding Technique Ourdata hiding technique extracts the samples of the EEG signalat the same time conventional video H264AVC decoding
6 BioMed Research International
Algorithm Embed a sample in low-motion sequencesInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) if PSKIP partition do(4) curblock = P16 times 16 partition type(5) switchMapping do(6) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(7) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(8) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(9) case 11 Quarter-Pixel Search Position = 0(10) end switch(11) end ifrarr until complete all partitions of the current macroblock(12) end forrarr until complete all samples
Algorithm 2 Algorithm of our data hiding encoding technique for low motion sequences
Referencevideo
frames
Read MV ofMB partitions
Extractbit pairs
Physiologicalsignaloutput
Video
Encoded videosequence
No
YesConcatenated
value
iquestallsamples
Figure 5 Sample extraction during the decoding process
process takes placeThis is an important feature at the time todo synchronized playback of both video and physiological-signal waveforms The sample extraction algorithm in ourtechnique catches and gathers sample data This extractionalgorithm illustrated in Figure 5 is repeated as many timesas needed to extract all the samples embedded in the videosequence To do this our data hiding technique interacts withthe H264AVC decoding process
The routine reads the motion vector (MV) of every mac-roblock partition and inputs the MV
119909and MV
119910components
into (4) to identify a binary combinationThe |MV|2 termsoutput the quarter-pixel search point values The routineobtains the bit-pair samples after identifying the 1198661 to 1198664values mapped into (2)
type isin
1198661 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
1198662 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198663 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198664 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
(4)
Algorithm 3 presents details of our proposed decodingalgorithm This decoding process extracts the physiologicalsamples from the encoded video sequences
4 Results
Theexperimental setup illustrated in Figure 6 included a PCan EEG database and a set of seven video test sequencesThe EEG samples were extracted from the CHB-MIT ScalpEEGDatabaseThis database is located at the PhysioBankdig-ital recordings (httpwwwphysionetorg)The experimentalsetup included 6 signal electrodes at 256 samples per secondand 12-bit sample resolution [1] The EEG samples generatedin one second were embedded into the first 30 frames of eachtest video sequence to establish synchronization
The implementation of encoding and decoding processesneeded the modification of the JM reference software version161 Table 3 shows the JM configuration A program wasdeveloped to provide and convert samples from the databaseto the encoder
The video test sequences had a CIF (352times288) resolutionand a 420 YUV format The name of the test sequences are
BioMed Research International 7
Algorithm Extract a sampleInput Embedded sample in a synchronous frameResult Data of the sample in bit-pairs according to (2)(1) for ExtractSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case (|MV
119909|2 = 1) amp (|MV
119910|2 = 0) Bit-pair = 00
(6) case (|MV119909|2 = 0) amp (|MV
119910|2 = a) Bit-pair = 01
(7) case (|MV119909|2 = 1) amp (|MV
119910|2 = 1) Bit-pair = 10
(8) case (|MV119909|2 = 0) amp (|MV
119910|2 = 0) Bit-pair = 11
(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 3 Algorithm of our data hiding decoding technique
Experimental setup
Video testsequences
Neonatal
EEG
samples
Storage
Transmission
ModifiedJM reference
H264AVC
09998
09989
09985
09973
09944
09913
0 1 2 3
Time (s)
Figure 6 Experimental setup for our synchronization data hiding scheme
Table 3 Configuration parameters for JM reference software
Parameter ResolutionProfile BaselineFrames 30Motion estimation algorithm Full searchRD optimization and rate control Disabled8 times 8 subblocks DisabledNumber of reference frames 1Quantization parameter (QP) 28
akiyo bridge-far carphone football foreman mobile andneonatal Neonatal is not considered a standard video testsequence Neonatal was introduced to match the context ofthe application example related to EEG seizures on neonatesThe selected coding structure of the bit stream is ldquoIPPP rdquo tohave an intraframe encoded in the first frame and interframesencoded in the remaining frames
Metrics to evaluate the effectiveness of our proposed tech-nique are video objective quality bit-rate difference embed-ding cost and perceptual quality of the image The peak
signal-to-noise ratio (PSNR) illustrated in (5) is an objectivequality metric to report video image degradation119872 and 119873are the height and the width of the video frame respectively119868119894119895 and 1198681198941198951015840 represent the original pixel and the processedpixel respectively [5] The PSNR
119884diff metric illustrated in(6) indicates how the luma (119884) samples impact the videoquality after embedding the physiological samples PSNR1015840
119884
represents the impact of luma samples generated by embed-ding the samples and PSNR
119884represents the impact of luma
samples generated with the original H264AVC encoder
PSNR = 10 logmax (1198681198941198952)
(1119872)sum
119872
119894sum
119873
119895[119868119894119895 minus 119868119894119895
1015840]
2 (5)
PSNR119884diff = PSNR
1015840
119884minus PSNR
119884
(6)
Equation (7) shows how to calculate the change of bit-rate (BRI) 119877 is the original bit-rate and 1198771015840 is the embeddingsamples bit-rate Equation (8) shows how to estimate theembedded cost Oe Ov is the data volume generated by theoriginal video coder DHe is the data volume generated byembedding the samples and EEGe is the embedded datavolume [7] The EEGe term refers to the data coming from
8 BioMed Research International
Table 4 Experimental results for seven video test sequences
Figure 7 PSNR luma difference values for neonatal video sequence
Football Mobile Foreman Carphone Neonatal Akiyo Bridge-far
Video test sequences
High motionMedium motionLow motion
00005
0010015
0020025
0030035
minus001minus0005PS
NR Y
diffe
renc
e (dB
)
Figure 8 PSNR luma difference for seven video test sequences
the EEG signals The perceptual quality provides an estima-tion of subjective quality of the image obtained by visualinspection
BRI = 1198771015840minus 119877
119877
times 100 (7)
Oe = DHe minusOv minus EEGeOv + EEGe
times 100 (8)
Figure 7 compares the PSNR119884difference between the
original and the embedding data encoding processes Thegraph shows a very small difference between PSNR
119884values
for the first 30 frames of the neonatal video sequence Thelargest increment in quality was 0187 dB This incrementappeared in the 29th frame The largest decrement in qualitywas minus0039 dB This decrement appeared in the 8th frame
Figure 8 demonstrates that the embedding process hasminor effect on the video objective quality of seven testsequences The 119910-axis represents the PSNR luma differencevalues expressed in decibels The positive values above aver-age 0011 dB and 0029 dB represent a very small improve-ment in video qualityThe negative values from minus0002 dB to
05
10152025303540
Bit-r
ate d
iffer
ence
()
Carphone Neonatal Akiyo Bridge-farFootball Mobile Foreman
Video test sequences
High motionMedium motionLow motion
Figure 9 Change of bit-rate for seven video test sequences
minus0007 represent a small degradation in video quality Thesedegradations and improvements are due to the lack of bit-rateconstraint
Embedding the EEG data into the video sequencesproduced increments of bit-rate for the seven video testsequences as illustrated in Figure 9 Our experimental setupincluded high motion medium motion and low motionvideo test sequences The amount motion of the neonatalvideo sequence was considered between medium and lowThe graph presents the neonatal sequence with gray colorto indicate that it is not a standard video test sequence Forthe high and medium motion sequences there are small bit-rate increments The bit-rate increments ranged from 045to 134 For the neonatal and low motion sequences thePSKIP block replacement occurred and resulted in bigger bit-rate incrementsThe bit-rate increments ranged from 56 to3658 However the changes in the bit-rate had no effect onthe video-waveform synchronization
Table 4 shows the results for the seven test sequences interms of embedded capacity modified macroblocks mod-ified motion vectors PSNR
119884difference bit-rate and the
embedded cost For the high medium motion and neonatalsequences the embedded cost ranged from minus01273 tominus26196 representing savings in data volume For the lowmotion sequences the embedded cost had an incrementin 158726 and 193017 These results indicate that ourdata hiding technique offers both an adequate efficiency forvideo-signal transmission and savings in the storage of highmedium and neonatal sequences
The inspection of subject quality indicated minimumvisual artifacts or distortion between original and data-hiding
BioMed Research International 9
(a) (b)
(c)
Figure 10 10th frame of neonatal video sequence (a) Original image (b) image using video original coding and (c) image with EEG data
images Figure 10 illustrates the perceptual quality of thevideo frames for the 10th frame of the neonatal sequenceThus data hiding generates no significant difference inquality from a human eye perception
Our technique will offer an adequate performance in thecontext of the application examples presented in Section 23The video sequences of these applications exhibit sufficientamount of motion In the neonatal-seizures application asufficient amount of motion is needed to identify clinicalmanifestations like epileptic attacks In the application aboutbreathing disorders respiratory and abnormal movementsare needed for accurate diagnosis In the application aboutcongestive heart failures the motion is associated with thephysical activities of the patient Finally the amount ofmotion is associated with the medical staff activity ratherthan the patient for the application example related to theimprovement of medical practices in operating rooms
5 Conclusions
The proposed data hiding technique was demonstrated tobe suitable for the medicine healthcare context Our tech-nique successfully embedded samples of six EEG signalsinto encoded video sequences with high medium and lowmotion Our technique also extracted the hidden samplesfrom the encoded video sequences without loss of informa-tion The implementation of our technique required simplertasks compared to other existing synchronization techniques(1) less number of encoders and decoders (2) no timestampsneeded (3) no software needed for synchronization of videoand signal streams and (4) higher data capacity compared
to other data hiding techniques especially for high motionsequences
The experimental results demonstrated minimum degra-dation in video quality and data savings in terms of storagetransmission The experimental results for high and mediummotion video test sequences ranged from minus0007 dB to0011 dB in PSNR luma difference from 04459 to 13446in the bit-rate difference and from minus26196 to minus01273in embedded cost The changes in PSNR
119884difference and
bit-rate resulted in both no impacts in video-waveform syn-chronization and minimum distortions in video quality Forstorage and transmission purposes the embedded cost forhigh andmediummotion video sequences represent savingsFor low motion video sequences the experimental resultsranged from ndash0003 dB to minus0002 dB in PSNR
119884difference
from 333143 to 36575 in the bit-rate difference and from158725 to 193017 in embedded cost The changes in bit-rate were higher compared to the high and medium videosequences
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported by Latin American and CaribbeanCollaborative ICT Research Federation (LACCIR) through aresearch grant (R1209LAC001)
10 BioMed Research International
References
[1] S Bhattacharyya A Roy D P Dogra et al ldquoSummarizationof neonatal video EEG for seizure and artifact detectionrdquoin Proceedings of the 3rd National Conference on ComputerVision Pattern Recognition Image Processing and Graphics(NCVPRIPG rsquo11) pp 134ndash137 Hubli India December 2011
[2] J Zhang Y Li and Y Wei ldquoUsing timestamp to realize audio-video synchronization in real-time streaming media transmis-sionrdquo in Proceedings of the International Conference on AudioLanguage and Image Processing (ICALIP 08) pp 1073ndash1076Shanghai China July 2008
[3] S K Kapotas E E Varsaki and A N Skodras ldquoData hidingin H 264 encoded video sequencesrdquo in Proceedings of the 9thIEEE Workshop on Multimedia Signal Processing (MMSP rsquo07)pp 373ndash376 October 2007
[4] X Li H Chen D Wang and X Qi ldquoAudio-video synchronouscoding based on mode selection in H264rdquo in Proceedings of the4th International Congress on Image and Signal Processing (CISPrsquo11) vol 1 pp 113ndash117 October 2011
[5] B Li and M-Q Shi ldquoAudio-video synchronization codingapproach based on H264AVCrdquo IEICE Electronics Express vol6 no 22 pp 1556ndash1561 2009
[6] H-M Chao C-M Hsu and S-G Miaou ldquoA data-hiding tech-nique with authentication integration and confidentiality forelectronic patient recordsrdquo IEEE Transactions on InformationTechnology in Biomedicine vol 6 no 1 pp 46ndash53 2002
[7] C-H Liu and O T-C Chen ldquoData hiding in inter and intraprediction modes of H264AVCrdquo in Proceedings of the IEEEInternational Symposium on Circuits and Systems (ISCAS rsquo08)pp 3025ndash3028 May 2008
[8] K Liao D Ye S Lian Z Guo and J Wang ldquoLightweightinformation hiding inH264AVC video streamrdquo in Proceedingsof the International Conference on Multimedia InformationNetworking and Security vol 1 pp 578ndash582 November 2009
[9] H Zhu R Wang and D Xu ldquoInformation hiding algorithmfor H264 based on the motion estimation of quarter-pixelrdquoin Proceedings of the 2nd International Conference on FutureComputer and Communication (ICFCC rsquo10) vol 1 pp V1-423ndashV1-427 IEEE Wuhan China May 2010
[10] Y Li H-X Chen and Y Zhao ldquoA new method of data hidingbased on H264 encoded video sequencesrdquo in Proceedings of theIEEE 10th International Conference on Signal Processing (ICSPrsquo10) pp 1833ndash1836 October 2010
[11] P Wang Z Zheng and L Li ldquoA video watermarking schemebased on motion vectors and mode selectionrdquo in Proceedings ofthe International Conference on Computer Science and SoftwareEngineering (CSSE 08) vol 5 pp 233ndash237 Wuhan ChinaDecember 2008
[12] K Banno and M H Kryger ldquoUse of polysomnography withsynchronized digital video recording to diagnose pediatric sleepbreathing disordersrdquoCanadianMedical Association Journal vol173 no 1 pp 28ndash30 2005
[13] H J G van den Berg-Emons J B J Bussmann H M MBalk and H J Stam ldquoValidity of ambulatory accelerometry toquantify physical activity in heart failurerdquo Scandinavian Journalof Rehabilitation Medicine vol 32 no 4 pp 187ndash192 2000
[14] M B Weinger D C Gonzales J Slagle and M Syeed ldquoVideocapture of clinical care to enhance patient safetyrdquo Quality andSafety in Health Care vol 13 no 2 pp 136ndash144 2004
Figure 4 The block diagram of our proposed data hiding technique based on H264AVC encoder
Algorithm Embed a sampleInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(6) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(7) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(8) case 11 Quarter-Pixel Search Position = 0(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 1 Algorithm of our data hiding encoding technique
the search point in each group with the minimum distortioncost In (3) 119869119894 is the Lagrangian rate distortion parameter
Our proposed technique illustrated in Figure 4 repeatsthe gray blocks until no samples are available The blocksin gray are the additional routines needed to implementour proposed data hiding algorithm The alternate pathexecutes the original H264AVCmotion estimation processAlgorithm 1 presents details of our data hiding algorithmfor encoding Our technique hides the signal samples in themotion vectors of each block partition located at a frameEach motion vectors hides two bits of a signal sample Ourtechnique hides no samples in the block types I4MB andI16MB due to their association with intraframe prediction
Our technique is also unable to hide samples into PSKIPblocks due to the lack of motion vectors
Our technique also incorporates an approach to over-come the data-hiding capacity limitation found in low-motion video sequences PSKIP blocks are the most com-mon block partition found in encoded low-motion videosequences A large number of PSKIP blocks limit the data hid-ing capacity of our technique due to few number of motionvectors found in the low-motion video sequence Thereforeour proposed technique forces the H264AVC encoder toreplace the PSKIP block by a P16 times 16 block partitionThis PSKIP replacement adds a motion vector to the datahiding capacity of our techniqueThePSKIP replacement alsocontributes to maintaining the synchronization between theencoded video and the physiological signals However thisreplacement may also result in an increment in the bit-rate ofthe sequence Algorithm 2 presents details of the algorithmfor low-motion sequences
32 Decoding Process of Our Data Hiding Technique Ourdata hiding technique extracts the samples of the EEG signalat the same time conventional video H264AVC decoding
6 BioMed Research International
Algorithm Embed a sample in low-motion sequencesInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) if PSKIP partition do(4) curblock = P16 times 16 partition type(5) switchMapping do(6) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(7) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(8) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(9) case 11 Quarter-Pixel Search Position = 0(10) end switch(11) end ifrarr until complete all partitions of the current macroblock(12) end forrarr until complete all samples
Algorithm 2 Algorithm of our data hiding encoding technique for low motion sequences
Referencevideo
frames
Read MV ofMB partitions
Extractbit pairs
Physiologicalsignaloutput
Video
Encoded videosequence
No
YesConcatenated
value
iquestallsamples
Figure 5 Sample extraction during the decoding process
process takes placeThis is an important feature at the time todo synchronized playback of both video and physiological-signal waveforms The sample extraction algorithm in ourtechnique catches and gathers sample data This extractionalgorithm illustrated in Figure 5 is repeated as many timesas needed to extract all the samples embedded in the videosequence To do this our data hiding technique interacts withthe H264AVC decoding process
The routine reads the motion vector (MV) of every mac-roblock partition and inputs the MV
119909and MV
119910components
into (4) to identify a binary combinationThe |MV|2 termsoutput the quarter-pixel search point values The routineobtains the bit-pair samples after identifying the 1198661 to 1198664values mapped into (2)
type isin
1198661 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
1198662 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198663 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198664 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
(4)
Algorithm 3 presents details of our proposed decodingalgorithm This decoding process extracts the physiologicalsamples from the encoded video sequences
4 Results
Theexperimental setup illustrated in Figure 6 included a PCan EEG database and a set of seven video test sequencesThe EEG samples were extracted from the CHB-MIT ScalpEEGDatabaseThis database is located at the PhysioBankdig-ital recordings (httpwwwphysionetorg)The experimentalsetup included 6 signal electrodes at 256 samples per secondand 12-bit sample resolution [1] The EEG samples generatedin one second were embedded into the first 30 frames of eachtest video sequence to establish synchronization
The implementation of encoding and decoding processesneeded the modification of the JM reference software version161 Table 3 shows the JM configuration A program wasdeveloped to provide and convert samples from the databaseto the encoder
The video test sequences had a CIF (352times288) resolutionand a 420 YUV format The name of the test sequences are
BioMed Research International 7
Algorithm Extract a sampleInput Embedded sample in a synchronous frameResult Data of the sample in bit-pairs according to (2)(1) for ExtractSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case (|MV
119909|2 = 1) amp (|MV
119910|2 = 0) Bit-pair = 00
(6) case (|MV119909|2 = 0) amp (|MV
119910|2 = a) Bit-pair = 01
(7) case (|MV119909|2 = 1) amp (|MV
119910|2 = 1) Bit-pair = 10
(8) case (|MV119909|2 = 0) amp (|MV
119910|2 = 0) Bit-pair = 11
(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 3 Algorithm of our data hiding decoding technique
Experimental setup
Video testsequences
Neonatal
EEG
samples
Storage
Transmission
ModifiedJM reference
H264AVC
09998
09989
09985
09973
09944
09913
0 1 2 3
Time (s)
Figure 6 Experimental setup for our synchronization data hiding scheme
Table 3 Configuration parameters for JM reference software
Parameter ResolutionProfile BaselineFrames 30Motion estimation algorithm Full searchRD optimization and rate control Disabled8 times 8 subblocks DisabledNumber of reference frames 1Quantization parameter (QP) 28
akiyo bridge-far carphone football foreman mobile andneonatal Neonatal is not considered a standard video testsequence Neonatal was introduced to match the context ofthe application example related to EEG seizures on neonatesThe selected coding structure of the bit stream is ldquoIPPP rdquo tohave an intraframe encoded in the first frame and interframesencoded in the remaining frames
Metrics to evaluate the effectiveness of our proposed tech-nique are video objective quality bit-rate difference embed-ding cost and perceptual quality of the image The peak
signal-to-noise ratio (PSNR) illustrated in (5) is an objectivequality metric to report video image degradation119872 and 119873are the height and the width of the video frame respectively119868119894119895 and 1198681198941198951015840 represent the original pixel and the processedpixel respectively [5] The PSNR
119884diff metric illustrated in(6) indicates how the luma (119884) samples impact the videoquality after embedding the physiological samples PSNR1015840
119884
represents the impact of luma samples generated by embed-ding the samples and PSNR
119884represents the impact of luma
samples generated with the original H264AVC encoder
PSNR = 10 logmax (1198681198941198952)
(1119872)sum
119872
119894sum
119873
119895[119868119894119895 minus 119868119894119895
1015840]
2 (5)
PSNR119884diff = PSNR
1015840
119884minus PSNR
119884
(6)
Equation (7) shows how to calculate the change of bit-rate (BRI) 119877 is the original bit-rate and 1198771015840 is the embeddingsamples bit-rate Equation (8) shows how to estimate theembedded cost Oe Ov is the data volume generated by theoriginal video coder DHe is the data volume generated byembedding the samples and EEGe is the embedded datavolume [7] The EEGe term refers to the data coming from
8 BioMed Research International
Table 4 Experimental results for seven video test sequences
Figure 7 PSNR luma difference values for neonatal video sequence
Football Mobile Foreman Carphone Neonatal Akiyo Bridge-far
Video test sequences
High motionMedium motionLow motion
00005
0010015
0020025
0030035
minus001minus0005PS
NR Y
diffe
renc
e (dB
)
Figure 8 PSNR luma difference for seven video test sequences
the EEG signals The perceptual quality provides an estima-tion of subjective quality of the image obtained by visualinspection
BRI = 1198771015840minus 119877
119877
times 100 (7)
Oe = DHe minusOv minus EEGeOv + EEGe
times 100 (8)
Figure 7 compares the PSNR119884difference between the
original and the embedding data encoding processes Thegraph shows a very small difference between PSNR
119884values
for the first 30 frames of the neonatal video sequence Thelargest increment in quality was 0187 dB This incrementappeared in the 29th frame The largest decrement in qualitywas minus0039 dB This decrement appeared in the 8th frame
Figure 8 demonstrates that the embedding process hasminor effect on the video objective quality of seven testsequences The 119910-axis represents the PSNR luma differencevalues expressed in decibels The positive values above aver-age 0011 dB and 0029 dB represent a very small improve-ment in video qualityThe negative values from minus0002 dB to
05
10152025303540
Bit-r
ate d
iffer
ence
()
Carphone Neonatal Akiyo Bridge-farFootball Mobile Foreman
Video test sequences
High motionMedium motionLow motion
Figure 9 Change of bit-rate for seven video test sequences
minus0007 represent a small degradation in video quality Thesedegradations and improvements are due to the lack of bit-rateconstraint
Embedding the EEG data into the video sequencesproduced increments of bit-rate for the seven video testsequences as illustrated in Figure 9 Our experimental setupincluded high motion medium motion and low motionvideo test sequences The amount motion of the neonatalvideo sequence was considered between medium and lowThe graph presents the neonatal sequence with gray colorto indicate that it is not a standard video test sequence Forthe high and medium motion sequences there are small bit-rate increments The bit-rate increments ranged from 045to 134 For the neonatal and low motion sequences thePSKIP block replacement occurred and resulted in bigger bit-rate incrementsThe bit-rate increments ranged from 56 to3658 However the changes in the bit-rate had no effect onthe video-waveform synchronization
Table 4 shows the results for the seven test sequences interms of embedded capacity modified macroblocks mod-ified motion vectors PSNR
119884difference bit-rate and the
embedded cost For the high medium motion and neonatalsequences the embedded cost ranged from minus01273 tominus26196 representing savings in data volume For the lowmotion sequences the embedded cost had an incrementin 158726 and 193017 These results indicate that ourdata hiding technique offers both an adequate efficiency forvideo-signal transmission and savings in the storage of highmedium and neonatal sequences
The inspection of subject quality indicated minimumvisual artifacts or distortion between original and data-hiding
BioMed Research International 9
(a) (b)
(c)
Figure 10 10th frame of neonatal video sequence (a) Original image (b) image using video original coding and (c) image with EEG data
images Figure 10 illustrates the perceptual quality of thevideo frames for the 10th frame of the neonatal sequenceThus data hiding generates no significant difference inquality from a human eye perception
Our technique will offer an adequate performance in thecontext of the application examples presented in Section 23The video sequences of these applications exhibit sufficientamount of motion In the neonatal-seizures application asufficient amount of motion is needed to identify clinicalmanifestations like epileptic attacks In the application aboutbreathing disorders respiratory and abnormal movementsare needed for accurate diagnosis In the application aboutcongestive heart failures the motion is associated with thephysical activities of the patient Finally the amount ofmotion is associated with the medical staff activity ratherthan the patient for the application example related to theimprovement of medical practices in operating rooms
5 Conclusions
The proposed data hiding technique was demonstrated tobe suitable for the medicine healthcare context Our tech-nique successfully embedded samples of six EEG signalsinto encoded video sequences with high medium and lowmotion Our technique also extracted the hidden samplesfrom the encoded video sequences without loss of informa-tion The implementation of our technique required simplertasks compared to other existing synchronization techniques(1) less number of encoders and decoders (2) no timestampsneeded (3) no software needed for synchronization of videoand signal streams and (4) higher data capacity compared
to other data hiding techniques especially for high motionsequences
The experimental results demonstrated minimum degra-dation in video quality and data savings in terms of storagetransmission The experimental results for high and mediummotion video test sequences ranged from minus0007 dB to0011 dB in PSNR luma difference from 04459 to 13446in the bit-rate difference and from minus26196 to minus01273in embedded cost The changes in PSNR
119884difference and
bit-rate resulted in both no impacts in video-waveform syn-chronization and minimum distortions in video quality Forstorage and transmission purposes the embedded cost forhigh andmediummotion video sequences represent savingsFor low motion video sequences the experimental resultsranged from ndash0003 dB to minus0002 dB in PSNR
119884difference
from 333143 to 36575 in the bit-rate difference and from158725 to 193017 in embedded cost The changes in bit-rate were higher compared to the high and medium videosequences
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported by Latin American and CaribbeanCollaborative ICT Research Federation (LACCIR) through aresearch grant (R1209LAC001)
10 BioMed Research International
References
[1] S Bhattacharyya A Roy D P Dogra et al ldquoSummarizationof neonatal video EEG for seizure and artifact detectionrdquoin Proceedings of the 3rd National Conference on ComputerVision Pattern Recognition Image Processing and Graphics(NCVPRIPG rsquo11) pp 134ndash137 Hubli India December 2011
[2] J Zhang Y Li and Y Wei ldquoUsing timestamp to realize audio-video synchronization in real-time streaming media transmis-sionrdquo in Proceedings of the International Conference on AudioLanguage and Image Processing (ICALIP 08) pp 1073ndash1076Shanghai China July 2008
[3] S K Kapotas E E Varsaki and A N Skodras ldquoData hidingin H 264 encoded video sequencesrdquo in Proceedings of the 9thIEEE Workshop on Multimedia Signal Processing (MMSP rsquo07)pp 373ndash376 October 2007
[4] X Li H Chen D Wang and X Qi ldquoAudio-video synchronouscoding based on mode selection in H264rdquo in Proceedings of the4th International Congress on Image and Signal Processing (CISPrsquo11) vol 1 pp 113ndash117 October 2011
[5] B Li and M-Q Shi ldquoAudio-video synchronization codingapproach based on H264AVCrdquo IEICE Electronics Express vol6 no 22 pp 1556ndash1561 2009
[6] H-M Chao C-M Hsu and S-G Miaou ldquoA data-hiding tech-nique with authentication integration and confidentiality forelectronic patient recordsrdquo IEEE Transactions on InformationTechnology in Biomedicine vol 6 no 1 pp 46ndash53 2002
[7] C-H Liu and O T-C Chen ldquoData hiding in inter and intraprediction modes of H264AVCrdquo in Proceedings of the IEEEInternational Symposium on Circuits and Systems (ISCAS rsquo08)pp 3025ndash3028 May 2008
[8] K Liao D Ye S Lian Z Guo and J Wang ldquoLightweightinformation hiding inH264AVC video streamrdquo in Proceedingsof the International Conference on Multimedia InformationNetworking and Security vol 1 pp 578ndash582 November 2009
[9] H Zhu R Wang and D Xu ldquoInformation hiding algorithmfor H264 based on the motion estimation of quarter-pixelrdquoin Proceedings of the 2nd International Conference on FutureComputer and Communication (ICFCC rsquo10) vol 1 pp V1-423ndashV1-427 IEEE Wuhan China May 2010
[10] Y Li H-X Chen and Y Zhao ldquoA new method of data hidingbased on H264 encoded video sequencesrdquo in Proceedings of theIEEE 10th International Conference on Signal Processing (ICSPrsquo10) pp 1833ndash1836 October 2010
[11] P Wang Z Zheng and L Li ldquoA video watermarking schemebased on motion vectors and mode selectionrdquo in Proceedings ofthe International Conference on Computer Science and SoftwareEngineering (CSSE 08) vol 5 pp 233ndash237 Wuhan ChinaDecember 2008
[12] K Banno and M H Kryger ldquoUse of polysomnography withsynchronized digital video recording to diagnose pediatric sleepbreathing disordersrdquoCanadianMedical Association Journal vol173 no 1 pp 28ndash30 2005
[13] H J G van den Berg-Emons J B J Bussmann H M MBalk and H J Stam ldquoValidity of ambulatory accelerometry toquantify physical activity in heart failurerdquo Scandinavian Journalof Rehabilitation Medicine vol 32 no 4 pp 187ndash192 2000
[14] M B Weinger D C Gonzales J Slagle and M Syeed ldquoVideocapture of clinical care to enhance patient safetyrdquo Quality andSafety in Health Care vol 13 no 2 pp 136ndash144 2004
Algorithm Embed a sample in low-motion sequencesInput Data of the sample in bit-pairs according to (2)Result Embedded sample in a frame(1) for NewSample do(2) curblockrarr current macroblock(3) if PSKIP partition do(4) curblock = P16 times 16 partition type(5) switchMapping do(6) case 00 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 5 6(7) case 01 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 1 2(8) case 10 Quarter-Pixel Search Position = 119894 | min(119869119894) 119894 isin 3 4 7 8(9) case 11 Quarter-Pixel Search Position = 0(10) end switch(11) end ifrarr until complete all partitions of the current macroblock(12) end forrarr until complete all samples
Algorithm 2 Algorithm of our data hiding encoding technique for low motion sequences
Referencevideo
frames
Read MV ofMB partitions
Extractbit pairs
Physiologicalsignaloutput
Video
Encoded videosequence
No
YesConcatenated
value
iquestallsamples
Figure 5 Sample extraction during the decoding process
process takes placeThis is an important feature at the time todo synchronized playback of both video and physiological-signal waveforms The sample extraction algorithm in ourtechnique catches and gathers sample data This extractionalgorithm illustrated in Figure 5 is repeated as many timesas needed to extract all the samples embedded in the videosequence To do this our data hiding technique interacts withthe H264AVC decoding process
The routine reads the motion vector (MV) of every mac-roblock partition and inputs the MV
119909and MV
119910components
into (4) to identify a binary combinationThe |MV|2 termsoutput the quarter-pixel search point values The routineobtains the bit-pair samples after identifying the 1198661 to 1198664values mapped into (2)
type isin
1198661 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
1198662 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198663 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 1) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 1)
1198664 if (100381610038161003816
1003816
MV119909
1003816
1003816
1003816
1003816
2 = 0) (1003816100381610038161003816
1003816
MV119910
1003816
1003816
1003816
1003816
1003816
2 = 0)
(4)
Algorithm 3 presents details of our proposed decodingalgorithm This decoding process extracts the physiologicalsamples from the encoded video sequences
4 Results
Theexperimental setup illustrated in Figure 6 included a PCan EEG database and a set of seven video test sequencesThe EEG samples were extracted from the CHB-MIT ScalpEEGDatabaseThis database is located at the PhysioBankdig-ital recordings (httpwwwphysionetorg)The experimentalsetup included 6 signal electrodes at 256 samples per secondand 12-bit sample resolution [1] The EEG samples generatedin one second were embedded into the first 30 frames of eachtest video sequence to establish synchronization
The implementation of encoding and decoding processesneeded the modification of the JM reference software version161 Table 3 shows the JM configuration A program wasdeveloped to provide and convert samples from the databaseto the encoder
The video test sequences had a CIF (352times288) resolutionand a 420 YUV format The name of the test sequences are
BioMed Research International 7
Algorithm Extract a sampleInput Embedded sample in a synchronous frameResult Data of the sample in bit-pairs according to (2)(1) for ExtractSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case (|MV
119909|2 = 1) amp (|MV
119910|2 = 0) Bit-pair = 00
(6) case (|MV119909|2 = 0) amp (|MV
119910|2 = a) Bit-pair = 01
(7) case (|MV119909|2 = 1) amp (|MV
119910|2 = 1) Bit-pair = 10
(8) case (|MV119909|2 = 0) amp (|MV
119910|2 = 0) Bit-pair = 11
(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 3 Algorithm of our data hiding decoding technique
Experimental setup
Video testsequences
Neonatal
EEG
samples
Storage
Transmission
ModifiedJM reference
H264AVC
09998
09989
09985
09973
09944
09913
0 1 2 3
Time (s)
Figure 6 Experimental setup for our synchronization data hiding scheme
Table 3 Configuration parameters for JM reference software
Parameter ResolutionProfile BaselineFrames 30Motion estimation algorithm Full searchRD optimization and rate control Disabled8 times 8 subblocks DisabledNumber of reference frames 1Quantization parameter (QP) 28
akiyo bridge-far carphone football foreman mobile andneonatal Neonatal is not considered a standard video testsequence Neonatal was introduced to match the context ofthe application example related to EEG seizures on neonatesThe selected coding structure of the bit stream is ldquoIPPP rdquo tohave an intraframe encoded in the first frame and interframesencoded in the remaining frames
Metrics to evaluate the effectiveness of our proposed tech-nique are video objective quality bit-rate difference embed-ding cost and perceptual quality of the image The peak
signal-to-noise ratio (PSNR) illustrated in (5) is an objectivequality metric to report video image degradation119872 and 119873are the height and the width of the video frame respectively119868119894119895 and 1198681198941198951015840 represent the original pixel and the processedpixel respectively [5] The PSNR
119884diff metric illustrated in(6) indicates how the luma (119884) samples impact the videoquality after embedding the physiological samples PSNR1015840
119884
represents the impact of luma samples generated by embed-ding the samples and PSNR
119884represents the impact of luma
samples generated with the original H264AVC encoder
PSNR = 10 logmax (1198681198941198952)
(1119872)sum
119872
119894sum
119873
119895[119868119894119895 minus 119868119894119895
1015840]
2 (5)
PSNR119884diff = PSNR
1015840
119884minus PSNR
119884
(6)
Equation (7) shows how to calculate the change of bit-rate (BRI) 119877 is the original bit-rate and 1198771015840 is the embeddingsamples bit-rate Equation (8) shows how to estimate theembedded cost Oe Ov is the data volume generated by theoriginal video coder DHe is the data volume generated byembedding the samples and EEGe is the embedded datavolume [7] The EEGe term refers to the data coming from
8 BioMed Research International
Table 4 Experimental results for seven video test sequences
Figure 7 PSNR luma difference values for neonatal video sequence
Football Mobile Foreman Carphone Neonatal Akiyo Bridge-far
Video test sequences
High motionMedium motionLow motion
00005
0010015
0020025
0030035
minus001minus0005PS
NR Y
diffe
renc
e (dB
)
Figure 8 PSNR luma difference for seven video test sequences
the EEG signals The perceptual quality provides an estima-tion of subjective quality of the image obtained by visualinspection
BRI = 1198771015840minus 119877
119877
times 100 (7)
Oe = DHe minusOv minus EEGeOv + EEGe
times 100 (8)
Figure 7 compares the PSNR119884difference between the
original and the embedding data encoding processes Thegraph shows a very small difference between PSNR
119884values
for the first 30 frames of the neonatal video sequence Thelargest increment in quality was 0187 dB This incrementappeared in the 29th frame The largest decrement in qualitywas minus0039 dB This decrement appeared in the 8th frame
Figure 8 demonstrates that the embedding process hasminor effect on the video objective quality of seven testsequences The 119910-axis represents the PSNR luma differencevalues expressed in decibels The positive values above aver-age 0011 dB and 0029 dB represent a very small improve-ment in video qualityThe negative values from minus0002 dB to
05
10152025303540
Bit-r
ate d
iffer
ence
()
Carphone Neonatal Akiyo Bridge-farFootball Mobile Foreman
Video test sequences
High motionMedium motionLow motion
Figure 9 Change of bit-rate for seven video test sequences
minus0007 represent a small degradation in video quality Thesedegradations and improvements are due to the lack of bit-rateconstraint
Embedding the EEG data into the video sequencesproduced increments of bit-rate for the seven video testsequences as illustrated in Figure 9 Our experimental setupincluded high motion medium motion and low motionvideo test sequences The amount motion of the neonatalvideo sequence was considered between medium and lowThe graph presents the neonatal sequence with gray colorto indicate that it is not a standard video test sequence Forthe high and medium motion sequences there are small bit-rate increments The bit-rate increments ranged from 045to 134 For the neonatal and low motion sequences thePSKIP block replacement occurred and resulted in bigger bit-rate incrementsThe bit-rate increments ranged from 56 to3658 However the changes in the bit-rate had no effect onthe video-waveform synchronization
Table 4 shows the results for the seven test sequences interms of embedded capacity modified macroblocks mod-ified motion vectors PSNR
119884difference bit-rate and the
embedded cost For the high medium motion and neonatalsequences the embedded cost ranged from minus01273 tominus26196 representing savings in data volume For the lowmotion sequences the embedded cost had an incrementin 158726 and 193017 These results indicate that ourdata hiding technique offers both an adequate efficiency forvideo-signal transmission and savings in the storage of highmedium and neonatal sequences
The inspection of subject quality indicated minimumvisual artifacts or distortion between original and data-hiding
BioMed Research International 9
(a) (b)
(c)
Figure 10 10th frame of neonatal video sequence (a) Original image (b) image using video original coding and (c) image with EEG data
images Figure 10 illustrates the perceptual quality of thevideo frames for the 10th frame of the neonatal sequenceThus data hiding generates no significant difference inquality from a human eye perception
Our technique will offer an adequate performance in thecontext of the application examples presented in Section 23The video sequences of these applications exhibit sufficientamount of motion In the neonatal-seizures application asufficient amount of motion is needed to identify clinicalmanifestations like epileptic attacks In the application aboutbreathing disorders respiratory and abnormal movementsare needed for accurate diagnosis In the application aboutcongestive heart failures the motion is associated with thephysical activities of the patient Finally the amount ofmotion is associated with the medical staff activity ratherthan the patient for the application example related to theimprovement of medical practices in operating rooms
5 Conclusions
The proposed data hiding technique was demonstrated tobe suitable for the medicine healthcare context Our tech-nique successfully embedded samples of six EEG signalsinto encoded video sequences with high medium and lowmotion Our technique also extracted the hidden samplesfrom the encoded video sequences without loss of informa-tion The implementation of our technique required simplertasks compared to other existing synchronization techniques(1) less number of encoders and decoders (2) no timestampsneeded (3) no software needed for synchronization of videoand signal streams and (4) higher data capacity compared
to other data hiding techniques especially for high motionsequences
The experimental results demonstrated minimum degra-dation in video quality and data savings in terms of storagetransmission The experimental results for high and mediummotion video test sequences ranged from minus0007 dB to0011 dB in PSNR luma difference from 04459 to 13446in the bit-rate difference and from minus26196 to minus01273in embedded cost The changes in PSNR
119884difference and
bit-rate resulted in both no impacts in video-waveform syn-chronization and minimum distortions in video quality Forstorage and transmission purposes the embedded cost forhigh andmediummotion video sequences represent savingsFor low motion video sequences the experimental resultsranged from ndash0003 dB to minus0002 dB in PSNR
119884difference
from 333143 to 36575 in the bit-rate difference and from158725 to 193017 in embedded cost The changes in bit-rate were higher compared to the high and medium videosequences
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported by Latin American and CaribbeanCollaborative ICT Research Federation (LACCIR) through aresearch grant (R1209LAC001)
10 BioMed Research International
References
[1] S Bhattacharyya A Roy D P Dogra et al ldquoSummarizationof neonatal video EEG for seizure and artifact detectionrdquoin Proceedings of the 3rd National Conference on ComputerVision Pattern Recognition Image Processing and Graphics(NCVPRIPG rsquo11) pp 134ndash137 Hubli India December 2011
[2] J Zhang Y Li and Y Wei ldquoUsing timestamp to realize audio-video synchronization in real-time streaming media transmis-sionrdquo in Proceedings of the International Conference on AudioLanguage and Image Processing (ICALIP 08) pp 1073ndash1076Shanghai China July 2008
[3] S K Kapotas E E Varsaki and A N Skodras ldquoData hidingin H 264 encoded video sequencesrdquo in Proceedings of the 9thIEEE Workshop on Multimedia Signal Processing (MMSP rsquo07)pp 373ndash376 October 2007
[4] X Li H Chen D Wang and X Qi ldquoAudio-video synchronouscoding based on mode selection in H264rdquo in Proceedings of the4th International Congress on Image and Signal Processing (CISPrsquo11) vol 1 pp 113ndash117 October 2011
[5] B Li and M-Q Shi ldquoAudio-video synchronization codingapproach based on H264AVCrdquo IEICE Electronics Express vol6 no 22 pp 1556ndash1561 2009
[6] H-M Chao C-M Hsu and S-G Miaou ldquoA data-hiding tech-nique with authentication integration and confidentiality forelectronic patient recordsrdquo IEEE Transactions on InformationTechnology in Biomedicine vol 6 no 1 pp 46ndash53 2002
[7] C-H Liu and O T-C Chen ldquoData hiding in inter and intraprediction modes of H264AVCrdquo in Proceedings of the IEEEInternational Symposium on Circuits and Systems (ISCAS rsquo08)pp 3025ndash3028 May 2008
[8] K Liao D Ye S Lian Z Guo and J Wang ldquoLightweightinformation hiding inH264AVC video streamrdquo in Proceedingsof the International Conference on Multimedia InformationNetworking and Security vol 1 pp 578ndash582 November 2009
[9] H Zhu R Wang and D Xu ldquoInformation hiding algorithmfor H264 based on the motion estimation of quarter-pixelrdquoin Proceedings of the 2nd International Conference on FutureComputer and Communication (ICFCC rsquo10) vol 1 pp V1-423ndashV1-427 IEEE Wuhan China May 2010
[10] Y Li H-X Chen and Y Zhao ldquoA new method of data hidingbased on H264 encoded video sequencesrdquo in Proceedings of theIEEE 10th International Conference on Signal Processing (ICSPrsquo10) pp 1833ndash1836 October 2010
[11] P Wang Z Zheng and L Li ldquoA video watermarking schemebased on motion vectors and mode selectionrdquo in Proceedings ofthe International Conference on Computer Science and SoftwareEngineering (CSSE 08) vol 5 pp 233ndash237 Wuhan ChinaDecember 2008
[12] K Banno and M H Kryger ldquoUse of polysomnography withsynchronized digital video recording to diagnose pediatric sleepbreathing disordersrdquoCanadianMedical Association Journal vol173 no 1 pp 28ndash30 2005
[13] H J G van den Berg-Emons J B J Bussmann H M MBalk and H J Stam ldquoValidity of ambulatory accelerometry toquantify physical activity in heart failurerdquo Scandinavian Journalof Rehabilitation Medicine vol 32 no 4 pp 187ndash192 2000
[14] M B Weinger D C Gonzales J Slagle and M Syeed ldquoVideocapture of clinical care to enhance patient safetyrdquo Quality andSafety in Health Care vol 13 no 2 pp 136ndash144 2004
Algorithm Extract a sampleInput Embedded sample in a synchronous frameResult Data of the sample in bit-pairs according to (2)(1) for ExtractSample do(2) curblockrarr current macroblock(3) forMacroblockPartition do(4) switchMapping do(5) case (|MV
119909|2 = 1) amp (|MV
119910|2 = 0) Bit-pair = 00
(6) case (|MV119909|2 = 0) amp (|MV
119910|2 = a) Bit-pair = 01
(7) case (|MV119909|2 = 1) amp (|MV
119910|2 = 1) Bit-pair = 10
(8) case (|MV119909|2 = 0) amp (|MV
119910|2 = 0) Bit-pair = 11
(9) end switch(10) end forrarr until complete all partitions of the current macroblock(11) end forrarr until complete all samples
Algorithm 3 Algorithm of our data hiding decoding technique
Experimental setup
Video testsequences
Neonatal
EEG
samples
Storage
Transmission
ModifiedJM reference
H264AVC
09998
09989
09985
09973
09944
09913
0 1 2 3
Time (s)
Figure 6 Experimental setup for our synchronization data hiding scheme
Table 3 Configuration parameters for JM reference software
Parameter ResolutionProfile BaselineFrames 30Motion estimation algorithm Full searchRD optimization and rate control Disabled8 times 8 subblocks DisabledNumber of reference frames 1Quantization parameter (QP) 28
akiyo bridge-far carphone football foreman mobile andneonatal Neonatal is not considered a standard video testsequence Neonatal was introduced to match the context ofthe application example related to EEG seizures on neonatesThe selected coding structure of the bit stream is ldquoIPPP rdquo tohave an intraframe encoded in the first frame and interframesencoded in the remaining frames
Metrics to evaluate the effectiveness of our proposed tech-nique are video objective quality bit-rate difference embed-ding cost and perceptual quality of the image The peak
signal-to-noise ratio (PSNR) illustrated in (5) is an objectivequality metric to report video image degradation119872 and 119873are the height and the width of the video frame respectively119868119894119895 and 1198681198941198951015840 represent the original pixel and the processedpixel respectively [5] The PSNR
119884diff metric illustrated in(6) indicates how the luma (119884) samples impact the videoquality after embedding the physiological samples PSNR1015840
119884
represents the impact of luma samples generated by embed-ding the samples and PSNR
119884represents the impact of luma
samples generated with the original H264AVC encoder
PSNR = 10 logmax (1198681198941198952)
(1119872)sum
119872
119894sum
119873
119895[119868119894119895 minus 119868119894119895
1015840]
2 (5)
PSNR119884diff = PSNR
1015840
119884minus PSNR
119884
(6)
Equation (7) shows how to calculate the change of bit-rate (BRI) 119877 is the original bit-rate and 1198771015840 is the embeddingsamples bit-rate Equation (8) shows how to estimate theembedded cost Oe Ov is the data volume generated by theoriginal video coder DHe is the data volume generated byembedding the samples and EEGe is the embedded datavolume [7] The EEGe term refers to the data coming from
8 BioMed Research International
Table 4 Experimental results for seven video test sequences
Figure 7 PSNR luma difference values for neonatal video sequence
Football Mobile Foreman Carphone Neonatal Akiyo Bridge-far
Video test sequences
High motionMedium motionLow motion
00005
0010015
0020025
0030035
minus001minus0005PS
NR Y
diffe
renc
e (dB
)
Figure 8 PSNR luma difference for seven video test sequences
the EEG signals The perceptual quality provides an estima-tion of subjective quality of the image obtained by visualinspection
BRI = 1198771015840minus 119877
119877
times 100 (7)
Oe = DHe minusOv minus EEGeOv + EEGe
times 100 (8)
Figure 7 compares the PSNR119884difference between the
original and the embedding data encoding processes Thegraph shows a very small difference between PSNR
119884values
for the first 30 frames of the neonatal video sequence Thelargest increment in quality was 0187 dB This incrementappeared in the 29th frame The largest decrement in qualitywas minus0039 dB This decrement appeared in the 8th frame
Figure 8 demonstrates that the embedding process hasminor effect on the video objective quality of seven testsequences The 119910-axis represents the PSNR luma differencevalues expressed in decibels The positive values above aver-age 0011 dB and 0029 dB represent a very small improve-ment in video qualityThe negative values from minus0002 dB to
05
10152025303540
Bit-r
ate d
iffer
ence
()
Carphone Neonatal Akiyo Bridge-farFootball Mobile Foreman
Video test sequences
High motionMedium motionLow motion
Figure 9 Change of bit-rate for seven video test sequences
minus0007 represent a small degradation in video quality Thesedegradations and improvements are due to the lack of bit-rateconstraint
Embedding the EEG data into the video sequencesproduced increments of bit-rate for the seven video testsequences as illustrated in Figure 9 Our experimental setupincluded high motion medium motion and low motionvideo test sequences The amount motion of the neonatalvideo sequence was considered between medium and lowThe graph presents the neonatal sequence with gray colorto indicate that it is not a standard video test sequence Forthe high and medium motion sequences there are small bit-rate increments The bit-rate increments ranged from 045to 134 For the neonatal and low motion sequences thePSKIP block replacement occurred and resulted in bigger bit-rate incrementsThe bit-rate increments ranged from 56 to3658 However the changes in the bit-rate had no effect onthe video-waveform synchronization
Table 4 shows the results for the seven test sequences interms of embedded capacity modified macroblocks mod-ified motion vectors PSNR
119884difference bit-rate and the
embedded cost For the high medium motion and neonatalsequences the embedded cost ranged from minus01273 tominus26196 representing savings in data volume For the lowmotion sequences the embedded cost had an incrementin 158726 and 193017 These results indicate that ourdata hiding technique offers both an adequate efficiency forvideo-signal transmission and savings in the storage of highmedium and neonatal sequences
The inspection of subject quality indicated minimumvisual artifacts or distortion between original and data-hiding
BioMed Research International 9
(a) (b)
(c)
Figure 10 10th frame of neonatal video sequence (a) Original image (b) image using video original coding and (c) image with EEG data
images Figure 10 illustrates the perceptual quality of thevideo frames for the 10th frame of the neonatal sequenceThus data hiding generates no significant difference inquality from a human eye perception
Our technique will offer an adequate performance in thecontext of the application examples presented in Section 23The video sequences of these applications exhibit sufficientamount of motion In the neonatal-seizures application asufficient amount of motion is needed to identify clinicalmanifestations like epileptic attacks In the application aboutbreathing disorders respiratory and abnormal movementsare needed for accurate diagnosis In the application aboutcongestive heart failures the motion is associated with thephysical activities of the patient Finally the amount ofmotion is associated with the medical staff activity ratherthan the patient for the application example related to theimprovement of medical practices in operating rooms
5 Conclusions
The proposed data hiding technique was demonstrated tobe suitable for the medicine healthcare context Our tech-nique successfully embedded samples of six EEG signalsinto encoded video sequences with high medium and lowmotion Our technique also extracted the hidden samplesfrom the encoded video sequences without loss of informa-tion The implementation of our technique required simplertasks compared to other existing synchronization techniques(1) less number of encoders and decoders (2) no timestampsneeded (3) no software needed for synchronization of videoand signal streams and (4) higher data capacity compared
to other data hiding techniques especially for high motionsequences
The experimental results demonstrated minimum degra-dation in video quality and data savings in terms of storagetransmission The experimental results for high and mediummotion video test sequences ranged from minus0007 dB to0011 dB in PSNR luma difference from 04459 to 13446in the bit-rate difference and from minus26196 to minus01273in embedded cost The changes in PSNR
119884difference and
bit-rate resulted in both no impacts in video-waveform syn-chronization and minimum distortions in video quality Forstorage and transmission purposes the embedded cost forhigh andmediummotion video sequences represent savingsFor low motion video sequences the experimental resultsranged from ndash0003 dB to minus0002 dB in PSNR
119884difference
from 333143 to 36575 in the bit-rate difference and from158725 to 193017 in embedded cost The changes in bit-rate were higher compared to the high and medium videosequences
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported by Latin American and CaribbeanCollaborative ICT Research Federation (LACCIR) through aresearch grant (R1209LAC001)
10 BioMed Research International
References
[1] S Bhattacharyya A Roy D P Dogra et al ldquoSummarizationof neonatal video EEG for seizure and artifact detectionrdquoin Proceedings of the 3rd National Conference on ComputerVision Pattern Recognition Image Processing and Graphics(NCVPRIPG rsquo11) pp 134ndash137 Hubli India December 2011
[2] J Zhang Y Li and Y Wei ldquoUsing timestamp to realize audio-video synchronization in real-time streaming media transmis-sionrdquo in Proceedings of the International Conference on AudioLanguage and Image Processing (ICALIP 08) pp 1073ndash1076Shanghai China July 2008
[3] S K Kapotas E E Varsaki and A N Skodras ldquoData hidingin H 264 encoded video sequencesrdquo in Proceedings of the 9thIEEE Workshop on Multimedia Signal Processing (MMSP rsquo07)pp 373ndash376 October 2007
[4] X Li H Chen D Wang and X Qi ldquoAudio-video synchronouscoding based on mode selection in H264rdquo in Proceedings of the4th International Congress on Image and Signal Processing (CISPrsquo11) vol 1 pp 113ndash117 October 2011
[5] B Li and M-Q Shi ldquoAudio-video synchronization codingapproach based on H264AVCrdquo IEICE Electronics Express vol6 no 22 pp 1556ndash1561 2009
[6] H-M Chao C-M Hsu and S-G Miaou ldquoA data-hiding tech-nique with authentication integration and confidentiality forelectronic patient recordsrdquo IEEE Transactions on InformationTechnology in Biomedicine vol 6 no 1 pp 46ndash53 2002
[7] C-H Liu and O T-C Chen ldquoData hiding in inter and intraprediction modes of H264AVCrdquo in Proceedings of the IEEEInternational Symposium on Circuits and Systems (ISCAS rsquo08)pp 3025ndash3028 May 2008
[8] K Liao D Ye S Lian Z Guo and J Wang ldquoLightweightinformation hiding inH264AVC video streamrdquo in Proceedingsof the International Conference on Multimedia InformationNetworking and Security vol 1 pp 578ndash582 November 2009
[9] H Zhu R Wang and D Xu ldquoInformation hiding algorithmfor H264 based on the motion estimation of quarter-pixelrdquoin Proceedings of the 2nd International Conference on FutureComputer and Communication (ICFCC rsquo10) vol 1 pp V1-423ndashV1-427 IEEE Wuhan China May 2010
[10] Y Li H-X Chen and Y Zhao ldquoA new method of data hidingbased on H264 encoded video sequencesrdquo in Proceedings of theIEEE 10th International Conference on Signal Processing (ICSPrsquo10) pp 1833ndash1836 October 2010
[11] P Wang Z Zheng and L Li ldquoA video watermarking schemebased on motion vectors and mode selectionrdquo in Proceedings ofthe International Conference on Computer Science and SoftwareEngineering (CSSE 08) vol 5 pp 233ndash237 Wuhan ChinaDecember 2008
[12] K Banno and M H Kryger ldquoUse of polysomnography withsynchronized digital video recording to diagnose pediatric sleepbreathing disordersrdquoCanadianMedical Association Journal vol173 no 1 pp 28ndash30 2005
[13] H J G van den Berg-Emons J B J Bussmann H M MBalk and H J Stam ldquoValidity of ambulatory accelerometry toquantify physical activity in heart failurerdquo Scandinavian Journalof Rehabilitation Medicine vol 32 no 4 pp 187ndash192 2000
[14] M B Weinger D C Gonzales J Slagle and M Syeed ldquoVideocapture of clinical care to enhance patient safetyrdquo Quality andSafety in Health Care vol 13 no 2 pp 136ndash144 2004
Figure 7 PSNR luma difference values for neonatal video sequence
Football Mobile Foreman Carphone Neonatal Akiyo Bridge-far
Video test sequences
High motionMedium motionLow motion
00005
0010015
0020025
0030035
minus001minus0005PS
NR Y
diffe
renc
e (dB
)
Figure 8 PSNR luma difference for seven video test sequences
the EEG signals The perceptual quality provides an estima-tion of subjective quality of the image obtained by visualinspection
BRI = 1198771015840minus 119877
119877
times 100 (7)
Oe = DHe minusOv minus EEGeOv + EEGe
times 100 (8)
Figure 7 compares the PSNR119884difference between the
original and the embedding data encoding processes Thegraph shows a very small difference between PSNR
119884values
for the first 30 frames of the neonatal video sequence Thelargest increment in quality was 0187 dB This incrementappeared in the 29th frame The largest decrement in qualitywas minus0039 dB This decrement appeared in the 8th frame
Figure 8 demonstrates that the embedding process hasminor effect on the video objective quality of seven testsequences The 119910-axis represents the PSNR luma differencevalues expressed in decibels The positive values above aver-age 0011 dB and 0029 dB represent a very small improve-ment in video qualityThe negative values from minus0002 dB to
05
10152025303540
Bit-r
ate d
iffer
ence
()
Carphone Neonatal Akiyo Bridge-farFootball Mobile Foreman
Video test sequences
High motionMedium motionLow motion
Figure 9 Change of bit-rate for seven video test sequences
minus0007 represent a small degradation in video quality Thesedegradations and improvements are due to the lack of bit-rateconstraint
Embedding the EEG data into the video sequencesproduced increments of bit-rate for the seven video testsequences as illustrated in Figure 9 Our experimental setupincluded high motion medium motion and low motionvideo test sequences The amount motion of the neonatalvideo sequence was considered between medium and lowThe graph presents the neonatal sequence with gray colorto indicate that it is not a standard video test sequence Forthe high and medium motion sequences there are small bit-rate increments The bit-rate increments ranged from 045to 134 For the neonatal and low motion sequences thePSKIP block replacement occurred and resulted in bigger bit-rate incrementsThe bit-rate increments ranged from 56 to3658 However the changes in the bit-rate had no effect onthe video-waveform synchronization
Table 4 shows the results for the seven test sequences interms of embedded capacity modified macroblocks mod-ified motion vectors PSNR
119884difference bit-rate and the
embedded cost For the high medium motion and neonatalsequences the embedded cost ranged from minus01273 tominus26196 representing savings in data volume For the lowmotion sequences the embedded cost had an incrementin 158726 and 193017 These results indicate that ourdata hiding technique offers both an adequate efficiency forvideo-signal transmission and savings in the storage of highmedium and neonatal sequences
The inspection of subject quality indicated minimumvisual artifacts or distortion between original and data-hiding
BioMed Research International 9
(a) (b)
(c)
Figure 10 10th frame of neonatal video sequence (a) Original image (b) image using video original coding and (c) image with EEG data
images Figure 10 illustrates the perceptual quality of thevideo frames for the 10th frame of the neonatal sequenceThus data hiding generates no significant difference inquality from a human eye perception
Our technique will offer an adequate performance in thecontext of the application examples presented in Section 23The video sequences of these applications exhibit sufficientamount of motion In the neonatal-seizures application asufficient amount of motion is needed to identify clinicalmanifestations like epileptic attacks In the application aboutbreathing disorders respiratory and abnormal movementsare needed for accurate diagnosis In the application aboutcongestive heart failures the motion is associated with thephysical activities of the patient Finally the amount ofmotion is associated with the medical staff activity ratherthan the patient for the application example related to theimprovement of medical practices in operating rooms
5 Conclusions
The proposed data hiding technique was demonstrated tobe suitable for the medicine healthcare context Our tech-nique successfully embedded samples of six EEG signalsinto encoded video sequences with high medium and lowmotion Our technique also extracted the hidden samplesfrom the encoded video sequences without loss of informa-tion The implementation of our technique required simplertasks compared to other existing synchronization techniques(1) less number of encoders and decoders (2) no timestampsneeded (3) no software needed for synchronization of videoand signal streams and (4) higher data capacity compared
to other data hiding techniques especially for high motionsequences
The experimental results demonstrated minimum degra-dation in video quality and data savings in terms of storagetransmission The experimental results for high and mediummotion video test sequences ranged from minus0007 dB to0011 dB in PSNR luma difference from 04459 to 13446in the bit-rate difference and from minus26196 to minus01273in embedded cost The changes in PSNR
119884difference and
bit-rate resulted in both no impacts in video-waveform syn-chronization and minimum distortions in video quality Forstorage and transmission purposes the embedded cost forhigh andmediummotion video sequences represent savingsFor low motion video sequences the experimental resultsranged from ndash0003 dB to minus0002 dB in PSNR
119884difference
from 333143 to 36575 in the bit-rate difference and from158725 to 193017 in embedded cost The changes in bit-rate were higher compared to the high and medium videosequences
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported by Latin American and CaribbeanCollaborative ICT Research Federation (LACCIR) through aresearch grant (R1209LAC001)
10 BioMed Research International
References
[1] S Bhattacharyya A Roy D P Dogra et al ldquoSummarizationof neonatal video EEG for seizure and artifact detectionrdquoin Proceedings of the 3rd National Conference on ComputerVision Pattern Recognition Image Processing and Graphics(NCVPRIPG rsquo11) pp 134ndash137 Hubli India December 2011
[2] J Zhang Y Li and Y Wei ldquoUsing timestamp to realize audio-video synchronization in real-time streaming media transmis-sionrdquo in Proceedings of the International Conference on AudioLanguage and Image Processing (ICALIP 08) pp 1073ndash1076Shanghai China July 2008
[3] S K Kapotas E E Varsaki and A N Skodras ldquoData hidingin H 264 encoded video sequencesrdquo in Proceedings of the 9thIEEE Workshop on Multimedia Signal Processing (MMSP rsquo07)pp 373ndash376 October 2007
[4] X Li H Chen D Wang and X Qi ldquoAudio-video synchronouscoding based on mode selection in H264rdquo in Proceedings of the4th International Congress on Image and Signal Processing (CISPrsquo11) vol 1 pp 113ndash117 October 2011
[5] B Li and M-Q Shi ldquoAudio-video synchronization codingapproach based on H264AVCrdquo IEICE Electronics Express vol6 no 22 pp 1556ndash1561 2009
[6] H-M Chao C-M Hsu and S-G Miaou ldquoA data-hiding tech-nique with authentication integration and confidentiality forelectronic patient recordsrdquo IEEE Transactions on InformationTechnology in Biomedicine vol 6 no 1 pp 46ndash53 2002
[7] C-H Liu and O T-C Chen ldquoData hiding in inter and intraprediction modes of H264AVCrdquo in Proceedings of the IEEEInternational Symposium on Circuits and Systems (ISCAS rsquo08)pp 3025ndash3028 May 2008
[8] K Liao D Ye S Lian Z Guo and J Wang ldquoLightweightinformation hiding inH264AVC video streamrdquo in Proceedingsof the International Conference on Multimedia InformationNetworking and Security vol 1 pp 578ndash582 November 2009
[9] H Zhu R Wang and D Xu ldquoInformation hiding algorithmfor H264 based on the motion estimation of quarter-pixelrdquoin Proceedings of the 2nd International Conference on FutureComputer and Communication (ICFCC rsquo10) vol 1 pp V1-423ndashV1-427 IEEE Wuhan China May 2010
[10] Y Li H-X Chen and Y Zhao ldquoA new method of data hidingbased on H264 encoded video sequencesrdquo in Proceedings of theIEEE 10th International Conference on Signal Processing (ICSPrsquo10) pp 1833ndash1836 October 2010
[11] P Wang Z Zheng and L Li ldquoA video watermarking schemebased on motion vectors and mode selectionrdquo in Proceedings ofthe International Conference on Computer Science and SoftwareEngineering (CSSE 08) vol 5 pp 233ndash237 Wuhan ChinaDecember 2008
[12] K Banno and M H Kryger ldquoUse of polysomnography withsynchronized digital video recording to diagnose pediatric sleepbreathing disordersrdquoCanadianMedical Association Journal vol173 no 1 pp 28ndash30 2005
[13] H J G van den Berg-Emons J B J Bussmann H M MBalk and H J Stam ldquoValidity of ambulatory accelerometry toquantify physical activity in heart failurerdquo Scandinavian Journalof Rehabilitation Medicine vol 32 no 4 pp 187ndash192 2000
[14] M B Weinger D C Gonzales J Slagle and M Syeed ldquoVideocapture of clinical care to enhance patient safetyrdquo Quality andSafety in Health Care vol 13 no 2 pp 136ndash144 2004
Figure 10 10th frame of neonatal video sequence (a) Original image (b) image using video original coding and (c) image with EEG data
images Figure 10 illustrates the perceptual quality of thevideo frames for the 10th frame of the neonatal sequenceThus data hiding generates no significant difference inquality from a human eye perception
Our technique will offer an adequate performance in thecontext of the application examples presented in Section 23The video sequences of these applications exhibit sufficientamount of motion In the neonatal-seizures application asufficient amount of motion is needed to identify clinicalmanifestations like epileptic attacks In the application aboutbreathing disorders respiratory and abnormal movementsare needed for accurate diagnosis In the application aboutcongestive heart failures the motion is associated with thephysical activities of the patient Finally the amount ofmotion is associated with the medical staff activity ratherthan the patient for the application example related to theimprovement of medical practices in operating rooms
5 Conclusions
The proposed data hiding technique was demonstrated tobe suitable for the medicine healthcare context Our tech-nique successfully embedded samples of six EEG signalsinto encoded video sequences with high medium and lowmotion Our technique also extracted the hidden samplesfrom the encoded video sequences without loss of informa-tion The implementation of our technique required simplertasks compared to other existing synchronization techniques(1) less number of encoders and decoders (2) no timestampsneeded (3) no software needed for synchronization of videoand signal streams and (4) higher data capacity compared
to other data hiding techniques especially for high motionsequences
The experimental results demonstrated minimum degra-dation in video quality and data savings in terms of storagetransmission The experimental results for high and mediummotion video test sequences ranged from minus0007 dB to0011 dB in PSNR luma difference from 04459 to 13446in the bit-rate difference and from minus26196 to minus01273in embedded cost The changes in PSNR
119884difference and
bit-rate resulted in both no impacts in video-waveform syn-chronization and minimum distortions in video quality Forstorage and transmission purposes the embedded cost forhigh andmediummotion video sequences represent savingsFor low motion video sequences the experimental resultsranged from ndash0003 dB to minus0002 dB in PSNR
119884difference
from 333143 to 36575 in the bit-rate difference and from158725 to 193017 in embedded cost The changes in bit-rate were higher compared to the high and medium videosequences
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported by Latin American and CaribbeanCollaborative ICT Research Federation (LACCIR) through aresearch grant (R1209LAC001)
10 BioMed Research International
References
[1] S Bhattacharyya A Roy D P Dogra et al ldquoSummarizationof neonatal video EEG for seizure and artifact detectionrdquoin Proceedings of the 3rd National Conference on ComputerVision Pattern Recognition Image Processing and Graphics(NCVPRIPG rsquo11) pp 134ndash137 Hubli India December 2011
[2] J Zhang Y Li and Y Wei ldquoUsing timestamp to realize audio-video synchronization in real-time streaming media transmis-sionrdquo in Proceedings of the International Conference on AudioLanguage and Image Processing (ICALIP 08) pp 1073ndash1076Shanghai China July 2008
[3] S K Kapotas E E Varsaki and A N Skodras ldquoData hidingin H 264 encoded video sequencesrdquo in Proceedings of the 9thIEEE Workshop on Multimedia Signal Processing (MMSP rsquo07)pp 373ndash376 October 2007
[4] X Li H Chen D Wang and X Qi ldquoAudio-video synchronouscoding based on mode selection in H264rdquo in Proceedings of the4th International Congress on Image and Signal Processing (CISPrsquo11) vol 1 pp 113ndash117 October 2011
[5] B Li and M-Q Shi ldquoAudio-video synchronization codingapproach based on H264AVCrdquo IEICE Electronics Express vol6 no 22 pp 1556ndash1561 2009
[6] H-M Chao C-M Hsu and S-G Miaou ldquoA data-hiding tech-nique with authentication integration and confidentiality forelectronic patient recordsrdquo IEEE Transactions on InformationTechnology in Biomedicine vol 6 no 1 pp 46ndash53 2002
[7] C-H Liu and O T-C Chen ldquoData hiding in inter and intraprediction modes of H264AVCrdquo in Proceedings of the IEEEInternational Symposium on Circuits and Systems (ISCAS rsquo08)pp 3025ndash3028 May 2008
[8] K Liao D Ye S Lian Z Guo and J Wang ldquoLightweightinformation hiding inH264AVC video streamrdquo in Proceedingsof the International Conference on Multimedia InformationNetworking and Security vol 1 pp 578ndash582 November 2009
[9] H Zhu R Wang and D Xu ldquoInformation hiding algorithmfor H264 based on the motion estimation of quarter-pixelrdquoin Proceedings of the 2nd International Conference on FutureComputer and Communication (ICFCC rsquo10) vol 1 pp V1-423ndashV1-427 IEEE Wuhan China May 2010
[10] Y Li H-X Chen and Y Zhao ldquoA new method of data hidingbased on H264 encoded video sequencesrdquo in Proceedings of theIEEE 10th International Conference on Signal Processing (ICSPrsquo10) pp 1833ndash1836 October 2010
[11] P Wang Z Zheng and L Li ldquoA video watermarking schemebased on motion vectors and mode selectionrdquo in Proceedings ofthe International Conference on Computer Science and SoftwareEngineering (CSSE 08) vol 5 pp 233ndash237 Wuhan ChinaDecember 2008
[12] K Banno and M H Kryger ldquoUse of polysomnography withsynchronized digital video recording to diagnose pediatric sleepbreathing disordersrdquoCanadianMedical Association Journal vol173 no 1 pp 28ndash30 2005
[13] H J G van den Berg-Emons J B J Bussmann H M MBalk and H J Stam ldquoValidity of ambulatory accelerometry toquantify physical activity in heart failurerdquo Scandinavian Journalof Rehabilitation Medicine vol 32 no 4 pp 187ndash192 2000
[14] M B Weinger D C Gonzales J Slagle and M Syeed ldquoVideocapture of clinical care to enhance patient safetyrdquo Quality andSafety in Health Care vol 13 no 2 pp 136ndash144 2004
[1] S Bhattacharyya A Roy D P Dogra et al ldquoSummarizationof neonatal video EEG for seizure and artifact detectionrdquoin Proceedings of the 3rd National Conference on ComputerVision Pattern Recognition Image Processing and Graphics(NCVPRIPG rsquo11) pp 134ndash137 Hubli India December 2011
[2] J Zhang Y Li and Y Wei ldquoUsing timestamp to realize audio-video synchronization in real-time streaming media transmis-sionrdquo in Proceedings of the International Conference on AudioLanguage and Image Processing (ICALIP 08) pp 1073ndash1076Shanghai China July 2008
[3] S K Kapotas E E Varsaki and A N Skodras ldquoData hidingin H 264 encoded video sequencesrdquo in Proceedings of the 9thIEEE Workshop on Multimedia Signal Processing (MMSP rsquo07)pp 373ndash376 October 2007
[4] X Li H Chen D Wang and X Qi ldquoAudio-video synchronouscoding based on mode selection in H264rdquo in Proceedings of the4th International Congress on Image and Signal Processing (CISPrsquo11) vol 1 pp 113ndash117 October 2011
[5] B Li and M-Q Shi ldquoAudio-video synchronization codingapproach based on H264AVCrdquo IEICE Electronics Express vol6 no 22 pp 1556ndash1561 2009
[6] H-M Chao C-M Hsu and S-G Miaou ldquoA data-hiding tech-nique with authentication integration and confidentiality forelectronic patient recordsrdquo IEEE Transactions on InformationTechnology in Biomedicine vol 6 no 1 pp 46ndash53 2002
[7] C-H Liu and O T-C Chen ldquoData hiding in inter and intraprediction modes of H264AVCrdquo in Proceedings of the IEEEInternational Symposium on Circuits and Systems (ISCAS rsquo08)pp 3025ndash3028 May 2008
[8] K Liao D Ye S Lian Z Guo and J Wang ldquoLightweightinformation hiding inH264AVC video streamrdquo in Proceedingsof the International Conference on Multimedia InformationNetworking and Security vol 1 pp 578ndash582 November 2009
[9] H Zhu R Wang and D Xu ldquoInformation hiding algorithmfor H264 based on the motion estimation of quarter-pixelrdquoin Proceedings of the 2nd International Conference on FutureComputer and Communication (ICFCC rsquo10) vol 1 pp V1-423ndashV1-427 IEEE Wuhan China May 2010
[10] Y Li H-X Chen and Y Zhao ldquoA new method of data hidingbased on H264 encoded video sequencesrdquo in Proceedings of theIEEE 10th International Conference on Signal Processing (ICSPrsquo10) pp 1833ndash1836 October 2010
[11] P Wang Z Zheng and L Li ldquoA video watermarking schemebased on motion vectors and mode selectionrdquo in Proceedings ofthe International Conference on Computer Science and SoftwareEngineering (CSSE 08) vol 5 pp 233ndash237 Wuhan ChinaDecember 2008
[12] K Banno and M H Kryger ldquoUse of polysomnography withsynchronized digital video recording to diagnose pediatric sleepbreathing disordersrdquoCanadianMedical Association Journal vol173 no 1 pp 28ndash30 2005
[13] H J G van den Berg-Emons J B J Bussmann H M MBalk and H J Stam ldquoValidity of ambulatory accelerometry toquantify physical activity in heart failurerdquo Scandinavian Journalof Rehabilitation Medicine vol 32 no 4 pp 187ndash192 2000
[14] M B Weinger D C Gonzales J Slagle and M Syeed ldquoVideocapture of clinical care to enhance patient safetyrdquo Quality andSafety in Health Care vol 13 no 2 pp 136ndash144 2004