Transform Domain based Video Steganography … Domain based Video Steganography Techniques ... such as text, still images, video images, audio or network protocol can be used as the

Transform Domain based VideoSteganography Techniques

Mritha Ramalingam1and Nor Ashidi Mat Isa2

1,2Imaging and Intelligent System Research Team (ISRT), School of Electrical and Electronic Engineering,Universiti Sains Malaysia, Engineering Campus, Nibong Tebal, Malaysia,

Email: [email protected], [email protected]

ABSTRACT: Steganography is quiet important when trustworthy and secret data communication are indeed over the Internettransmission. Steganography is the art of camouflaging the secret data inside the carriers such as text, image, audio, videoor protocols. The secret data could be embedded in spatial or transform domains of the carrier or cover-medium. Thispaper discusses different steganography techniques of data hiding in transform domain of cover-medium. The problemsencountered and advantages of the data hiding methods are illustrated. With this fundamental study, as a result, the researchfocuses on the design of new transform domain based video steganography techniques using affine transforms, wavelettransforms and Markov models.

Keywords: video steganography, transform domain, data hiding, information security

1. INTRODUCTIONIn recent years, due to the technological advancements(Chen et al. 2011) in the open milieu like Internet, dataare at greater risk. Huge amount of information areconstantly transmitted over unsecured communicationchannels. However, the progress of Internetcommunication has highlighted the issue ofinformation security. Despite cryptography,steganography plays a vital role in information security.Cryptography keeps the messages secret andsteganography keep the mere existence of the messagesecret. But, in modern digital secret communicationsystems, the cryptography is combined withsteganography to further strengthen the secrecy.

Steganography is the science of covertcommunication. Steganography is the art of concealingsecret information in a carrier. Different digital mediasuch as text, still images, video images, audio or networkprotocol can be used as the carrier of the secret messagesin most of the steganography techniques (Anderson andSiome, 2008). The other digital technology,watermarking is also closely related to steganographybut watermarking is concerned about the protection ofthe ownership of intellectual properties.

Many steganography methods involve secretmessage hiding in spatial and transform domains of acover-medium. Spatial domain-based steganographyincludes the direct replacement of the pixel values of

the cover-medium. Transform domain-basedsteganography techniques involves the manipulation ofthe pixels of the cover-medium, and then the data arehidden in the transformed values. Hence, transformdomain-based embedding techniques offer a higherdegree of robustness than spatial domain methods(Adnan and Wafaa, 2010).

Moreover, using video as a carrier of secretmessage resolves the issues related to embeddingcapacity and security. Researchers have proposedvarious video steganography algorithms to achievesecret digital communication by addressing issues suchas computation and processing time. The principle ofvideo steganography in hiding and transmitting thesecret messages definitely favors various security-related applications. Thus, this paper is intended toprovide the review of various state-of-the-art videosteganography methods based on transforms domain.

This paper is organized into following sections.Section 2 presents the basic principle of a steganographymodel. Section 3 discusses different transform domain-based video steganography techniques. The sectiondiscusses different methods used in transforms domain.Finally, conclusion is drawn in section 4.

2. BASIC STEGANOGRAPHY MODELSteganography refers to the process of hiding data inanother data (Zielinska et al. 2014). Figure 1 shows

IJCSIE, Vol. 7, No. 1, June 2016

40 � IJCSIE

the basic steganography model. The cover-medium isthe carrier into which the secret message is embedded.The stego-medium is the output of the steganographymodel containing hidden message.

The secret message is embedded in the cover-medium using the steganography algorithm. Thesteganography algorithm tr ies to maintain theproperties of the cover-medium making the changeshardly noticeable by the human eye. In somesteganography models, a secret key is used in thesteganography process for hiding the secret message.The secret key is kept confidential among the senderand the receiver so that the intended receiver retrievesthe embedded message using it.

3. VIDEO STEGANOGRAPHYMECHANISMS BASED ON TRANSFORMDOMAIN

This section provides an in-depth review of thechallenges faced by video steganography mechanismsbased on the transform domain. The pros and cons ofdifferent video steganography methods based on thetransform domain are presented in this section. Theprincipal objective of video steganography is to hide asecret message without diminishing the visual quality,structure, and content of the video file.

Video steganography is performed in the spatialand transform domains. The spatial domain algorithmstraightforwardly embeds messages in the covermedium with no visual modifications. The outcome ofspatial domain algorithms is beneficial in terms ofcapacity. The transform domain algorithm embeds thesecret data in the transformed coefficients. Transformdomain techniques have the advantage of higherstability than spatial domain techniques (Mandal 2012).

The main challenge of video steganography is toembed the message in a way that could prevent theexposure of hidden messages in videos. A recent videosteganography method proposed by Tamer (2012) hidedata in MPEG video files using multivariate regression

and flexible macro block ordering. Tamer (2012)presented two data hiding approaches to address theissues related to HVS. The first approach modulatedthe quantization of constant bit-rate video, and thesecond approach used flexible macro block orderingto hide data. Although the methods had high predictionaccuracy using MPEG videos, the payload wasrestricted and interfered in the bit rate of the codedvideo, thereby affecting the video quality. Therobustness of the proposed compressed MPEG videoagainst channel bit errors and packet losses needadditional focus.

Transform domain steganography techniques arepreferable to spatial domain techniques. Varioustransformations are performed on the video for datahiding in transform domain techniques (Bin et al.2011). Yueyun (2007) proposed an information hidingmethod based on the frequency domain for MPEGvideo. Hidden data are recovered without using theoriginal MPEG video data in the proposed method.This technique works in the frequency domain only. Ithas the advantages of low complexity and low visualdistortion for covert communication applications.However, it suffers from low payload capacity.

Similarly, Aly (2011) hid data in compressedvideos. Data hidden in images and unrefined videos inthe spatial or transformed domain are susceptible tosteganalysis. Similarly, the motion vectors are intendedto encode and reconstruct both the forward P-frameand B-frames in compressed videos. The secretmessage bit stream is inserted in the LSB of bothelements of the applicant motion vectors. The optionof the applicant subset of these motion vectors is basedon their related macro block prediction error, which isdifferent from the approaches based on the motionvector attributes such as the magnitude and phase angle.The robustness is achieved using the greedy adaptivethreshold to speed up the search of every frame, therebypreserving a low prediction error level.

Hamid et al. (2009) proposed a digitalsteganography technique based on bit-plane complexitysegmentation (BPCS) for data embedding. BPCSachieves high inserting rates with low distortion. BPCSworks on the principle that noise like regions in imagesof bit planes are substituted with noise like secret data.This approach formulates highly secured data hiddenusing a selected frame in an MPEG video. The reasonfor the selected frame in the video cover is the largeamount of single frames within a period, which

Figure 1: Basic steganography model

Secret message Steganographyalgorithm

Cover-medium Stego-medium

Transform domain based video steganography techniques � 41

consecutively resolve the difficulty of hiding a largequantity of data. However, the method encountersdistortions by using digital video as a cover to hidedata.

Deepika et al. (2012) examined the data hidingprocess in natural sequences of multiple groups ofimages, and the outcomes are forecasted. The methodinvolved minimal distortions in building the video andleast overhead on the compressed video size. Thelossless compression quantization is less beneficialalthough the clarity of the video sequences ismaintained. As a solution to the issues related to theclarity of video files, a steganography system usingthe YCbCr color space for data hiding was proposedby Cheddad et al. (2008). YCbCr utilizes the benefitof human color-response features for data embedding.The proposed method with enhanced YCbCr colorspace has achieved better results than the F5 and S-Tools. Although the enhanced YCbCr tool outperformsother existing tools, it still requires an additionalintention especially in any rotational distortion action.

Qiang and Thomas (2001) presented a transformdomain approach that hides data and detects intrudersby using an optimum detection structure in contrast tothe spatial domain. The statistical behavior of sub-bandcoefficients is modeled by the generalized Gaussiandistribution. The structure of the optimum detection isbuilt, and the performance of the exact asymptoticdetection is evaluated to provide improved capacity.

Most transform domain techniques are performedusing different transformation functions such as affinetransform (Shen et al. 2012), discrete Fourier transform(DFT; McKeon 2007), discrete cosine transform (DCT;Nedal et al. 2009), and wavelet transforms such asDWT (Ashok et al. 2012) and IWT (Jayasudha 2012).However, affine transform, wavelet transform, andMarkov mechanisms are the most standardsteganographic techniques for embedding a secretmessage in high-quality video, whereas the HVS isunable to predict the hidden message in the cover-video.

3.1. Discrete Fourier TransformDFT is used to forward an image/video frame fromthe spatial domain to the frequency domain. It dealswith the finite amount of frequency data coefficients.Song et al. (2002) embedded data in the transformdomain using DFT coefficients. The method uses asystematic property of DFT for data hiding. The author

achieved high embedding capacity, but the quality ofthe carrier is degraded. McKeon (2007) used atechnique called strange Fourier steganography to hidedata in movies. This technique enabled the sender andthe receiver to have secured data communicationwithout being concerned whether a third party mightdetect the hidden data in the movie file. The DFT-basedsteganography methods face computationalinconvenience. Thus, other functions discussed in thenext section are used to perform the transformationsin video steganography.

3.2. Discrete Cosine TransformDCT is a function based on the transform domainsimilar to DFT. It is closely related to DFT but offers ahigher energy compaction property than DFT fornatural images (Singh and Siddiqui 2012). Thedifference between DCT and DFT is that the formeruses only real numbers, whereas the latter uses complexnumbers. DCT is a popular signal transformationmethod using cosine functions of different frequencies.It transforms the image from the spatial domain to thefrequency domain. It separates the image into spectralsub-bands with respect to its visual quality, that is, high-frequency, middle-frequency, and low-frequencycomponents for data embedding. Ekta et al. (2010)performed the data hiding approach using DCT andLSB techniques. The proposed DCT-basedsteganography embeds the text message in the LSB ofthe DCT coefficient of a digital picture. The dataembedding capacity of the system is limited althoughthe method achieved better quality.

Veerdeep et al. (2013) explained various digitalsteganography techniques of hiding data in the spatial,transform, and compression domains. The authors hiddata using DCT coefficients and vector quantizationof images to increase the quality. DCT facilitates theprocess of partitioning the image into sections orspectral sub-bands of reverse value in accordance withvisual quality. The most dominant quantizationtechnique used for the image or video compression isvector quantization (VQ). However, a trade-off existsbetween quality and embedding capacity. Similarly,Chang et al. (2009) created a data embedding techniqueto implant secrets into VQ directories by using theChinese remainder theorem. The thrashing capabilityis flexible, which corresponds to the predeterminedmajor constraints. A reversible method is introducedinto the system using this technique. Thus, the uniqueVQ directories can be reinstated after retrieving secret

42 � IJCSIE

repossession. The theorem embeds only a restrictedcapacity of the secret message.

Primechaev et al. (2007) addressed the problemof moving object segmentation or scene changedetection in DCT and sub-bands/wavelets. Theresearchers also intended to evaluate change detectionapproaches in compressed domain standards usingDCT features. Scene change detection using DCTcoefficients are used to detect only necessary regionsof interest for subsequent image-processing steps.However, in considering multi-image detection, thefaster-moving character probably results in lowerdetection accuracy. Hossein et al. (2012) used DCTcoefficients consecutively to hide secret data in theLSBs of pixel values. All of the DCT coefficients aremanipulated sequentially to hide secret data in theLSBs of pixel values of the cover. DCT values are notchanged directly, and only the middle frequencies arereplaced with secret data bits. This combination ofDCT coefficients and LSB has improved the securityof embedded data in video steganography.

Daniela et al. (2007) presented a DCT-based videosteganography to transfer subtitle data from the encoderto the decoder. The advantage of the method is thatthe scheme obtained an imperceptible decrease inpicture quality. However, the DCT coefficients shouldbe intact before being quantized. Thus, the totalcapacity of the hidden channel should be improved.Quantization is the process of restricting discrete valuesfrom a comparatively large or continuous set of realvalues to a comparatively small discrete set integer.As a solution, the DCT coefficients are quantizedbefore the data-hiding process (Linjie et al. 2014). Theauthors embedded data by uniformly spreading theminto quantized DCT coefficients of all possiblemagnitudes. The average modifications of first-orderand second-order statistics were possibly minimized,especially in the small coefficient that led to lessstatistical detectability. The method achieved bettersecurity.

Nedal et al. (2009) utilized the principle of SuperSmash Bros-4 (SSB-4), an advanced technique in aseries of actions. The method works by embeddingmessage bits in the fourth bit of the coefficients of atransform domain, such as the DCT and wavelet, of acarrier. The proposed technique employs the idea ofthe SSB-4 technique in modifying the other bits (i.e.,first, second, third, and/or fourth) to obtain theminimum variation between the original and modified

coefficients. The method achieved better embeddingrate and minimum distortion. Alternatively, Xiong etal. (2013) used DCT coefficient-based data hiding forwireless sensor network communication. The numberof most significant data bits embedded into the DCTcoefficients of the cover image is based on the DCTcoefficients to secure secret data. Based on theproposed energy consumption and rate distortionmodel, resource allocation is optimized with limitedenergy and bandwidth. Furthermore, other transformdomain-based functions, which are the focus of thisstudy, are described in the next section.

3.3. Affine TransformAffine transformation is a method that maintains thecollinearity and proportions of distances, such as scale,translation, and rotation, in any object with a linesegment. Affine transformation is expressed byEquation (1) as follows:

,X a b X e

Y c d Y f

�� � � � � � � �� �� � � � � � � ��� � � � � � � �

(1)

where (X, Y) is the pixel value of the original imageand (X� , Y�) is the pixel value of the image after affinetransformation. Affine transformation is achieved byassigning different values to the six parameters a, b, c,d, e, and f. For example, if the parameters a = d = 0,b = c = 0, and e = f = 0, then a scale transformation isobtained; and if a = d = cos f and b = c = –sin f, thenrotation and translation are obtained. No variation inthe image content through affine transformation wasobserved, but some changes in pixel values wereobserved when compared with the original image.Based on this feature, the histogram of the originalimage is maintained (Shen et al. 2012).

A simple type of affine transformation is appliedto the image and re-sampled using interpolation.Typical interpolation methods include nearest neighborinterpolation, bilinear interpolation, and cubicconvolution interpolation (Ranjeet et al. 2013). Figure2 shows that most of the image pixel values havechanged after interpolation.

The encoding process by Loay and Ahmed (2011)hides the secret message initially by converting themessage into coefficients of affine transformation.Then, the coordinates of each pixel of a selected objectin the cover image are recomputed using affinetransformation.

Transform domain based video steganography techniques � 43

Integer DCT is a proper domain for steganographybecause it is invertible and lossless. Xianhua et al.(2012) proposed a steganography method based onaffine transformation in combination with integer DCT.However, the change in DCT coefficients degrades theLaplacian shape, such as delivery capacity. Affinetransformation recovers the Laplacian shape, such asallocation of the integer DCT coefficients, to guaranteethe security of the method. Affine transformationprovides good security in transmitting a visually andtypically imperceptible stego-image even with a largepayload. Wavelet transforms domain-basedsteganography methods are discussed in the nextsection.

3.4. Wavelet TransformsWavelet transform domain-based steganographymethods include functions, such as DWT and IWT.Various steganography methods based on DWT andIWT are discussed in this section. Wavelet transformdelineates a multi-resolution decomposition of animage of video frames into a set of wavelet basisfunctions.

A video-based steganographic system wasproposed by Klimis et al. (2002) for multiuse messagebinding. The method embeds data into the mostsignificant wavelet coefficient to provide invisibilityand resistance against lossy transmission or otherdistortions. An absolute practical methodology thatminimized additive distortions in steganography wasestablished by Filler et al. (2011). The proponents useda novel syndrome coding scheme based on dualconvolution codes equipped with the Viterbi algorithmfor data embedding and achieved a fast and versatilesolution.

Pankajakshan and Ho (2007) presented a waveletdomain-based video steganalysis method. Theyinvestigated how the temporal correlation in theneighboring frames of an MPEG video sequence couldbe utilized to improve the performance of video

steganalysis. As a result, the method achieved a betterdata retrieval rate by utilizing temporal correlation.

3.4.1. Discrete Wavelet TransformEffective DWT is simply a small wave that has itsenergy focused on time to provide a means for theanalysis of data communication and nonstationary ortime-varying phenomena. A signal system is built withtwo metrics, one having a binary sum and the otherhaving a coefficient with two indices. The position ofcoefficients is called the DWT of a signal. The DWTpartitions the signal into high-frequency and low-frequency sections. The pixel sums represent the low-frequency part of the original image, denoted by theletter L. Meanwhile, the pixel differences represent thehigh-frequency part of the original image, denoted bythe letter H (Barve et al. 2012). The high-frequencysection comprises information about the edge elements.Meanwhile, the low-frequency section is partitionedagain into high-frequency and low-frequencysubsections. The high-frequency elements are typicallyused for steganography because HVS is less perceptiveto changes in edges.

DWT using the pixel mapping method of Souvikand Gautam (2010) resulted in a high-capacitysteganography technique. In 2D applications, for eachlevel of decomposition, DWT is initially conducted inthe vertical direction, followed by DWT in thehorizontal direction. The three-phase decompositionis depicted in Figure 3. As shown in the figure, afterthe first level of decomposition, four sub-bands,namely, LL1, LH1, HL1, and HH1, are obtained. For eachsuccessive level of decomposition, the LL sub-bandof the prior level is used as the input. For the next levelof decomposition, DWT decomposed the LL1 band intofour sub-bands, namely, LL2, LH2, HL2, and HH2

(Cláudio and Jacob 2005).

Forward DWT (Naoya et al. 2012) is appropriatefor prediction of the spot where secret data are hiddenin the cover medium using the potential space–frequency localization feature. More specifically, thisfeature allows the utilization of data embedding suchthat, if a DWT coefficient changes, then only the areacorresponding to that coefficient changes. Embeddingsecret data in the lower-frequency sub-bands (LLX)degrades the image considerably because most of theimage power is stored in these sub-bands. Conversely,embedding secret data in the low-frequency sub-bandscould enhance robustness extensively. The edges and

Figure 2: Pixel value changes in affine transformation.

Actualimage

Interpolatedimage

Interpolation ofimage

44 � IJCSIE

textures of the image and the human eye do nottypically detect changes in the high-frequency sub-bands (HHX), which allows secret data to be embeddedwithout being identified by HVS.

In contrast to the space domain approaches, datahiding in the frequency domain was conducted by Poand Hung (2006). The secret messages are embeddedin the high-frequency DWT coefficients. The DWTcoefficients in the low-frequency sub-bands areunmodified to enhance quality. Some basicmathematical operations are conducted on the secretmessage before embedding. The method was able toachieve better security.

The objective of many DWT-based algorithms isto achieve acceptable performance in terms ofimperceptibility and robustness, as well as to embedthe secret image in middle-frequency sub-bands (LHX)or (HLX) and (HHX) (Ashok et al. 2012). Meanwhile,Lin and Hing (2006) used the one third and roundingmethods for embedding data in wavelet coefficients toretain the co-occurrence matrix of the waveletcoefficient. The researchers hid the secret data in imageblocks with distortions based on the bit planecomplexity of each block. The method did not degradethe co-occurrence matrix of the wavelet coefficient.

Hongmei et al. (2011) presented a blind video data-hiding algorithm in the DWT domain by embeddingmultiple information bits into uncompressed videosequences. The method explicitly excluded the LL sub-

band coefficients and embedded data in the LL sub-band for better invisibility and robustness. The methodachieved better imperceptibility.

Hamsatvani (2012) presented DWT-based videosteganography. The method hides an image in videostreams using DWT and singular value decomposition.In this algorithm, the secret image is not embeddeddirectly in the wavelet coefficients but in the elementsof singular values of the DWT sub-bands of the videosequence. The method achieved better robustness andimperceptibility. Moreover, the DWT-basedsteganography technique of Seyedi et al. (2011)transforms the object in the wavelet domain, processesthe coefficients, and then conducts inverse wavelettransform to represent the actual format of the stego-medium. This DWT-based steganography methodembeds one to three images into another carrier image.

Kamstra et al. (2005) examined the high-capacitylossless data-embedding techniques that hide largeamounts of data into digital images (or video) in sucha manner that the unique image could not be recreatedfrom the watermarked image. Earlier studies havereported that steganography tools, for instance, the S-Tools, outperform other tools in thrashing data in thespatial region. The results show that these toolsoutperform the set of procedures for enhancing thesteganographic systems. However, in this method,enhanced presentation in recovering concealed data isobserved after relevant image-processing attacks in theform of additive artificial noise.

Vijay and Dinesh (2010) combined DWT and DCTto conduct steganography. First, the proposed schemeencodes the secret data by applying DCT. Then, theimage characteristics are extracted from the carrier andDWT is applied. Using two different secret keys, theencoded features of DCT coefficients are hidden in thecover image using the DWT function. Thesteganography scheme of Anjali et al. (2014) resultedin a high-capacity data-hiding method withoutsacrificing the quality of the cover image. DWT-baseddigital image steganography algorithm uses the strengthof wavelet transform domains to obtain furtherinvisibility and robustness. However, one of the maindisadvantages of the method is its inability to protectthe image from distortions.

3.4.2. Integer Wavelet TransformIWT prevents the floating point precision problems ofthe wavelet filter. IWT utilizes the spatial and temporalcorrelations in and between video frames to minimize

LL3 HL3

LH3 HH3

HL2

HL1

LH2 HH2

LH1 HH1

Figure 3: Three-phase decomposition using DWT

Transform domain based video steganography techniques � 45

the embedding noise present in video (Lakshmi et al.2012). Another advantage of the wavelet function isthat it supports multiple resolutions. Lakshmi et al.(2012) proposed a steganography system that utilizesIWT in the cover image to obtain the stego-image. Theproposed method obtains a stego-image with highcapacity and security. Time complexity in executionis high because selections are devised to determine themost appropriate component.

Blind video steganography using IWT wasproposed by Abbass et al. (2006). The method increasesthe hiding capacity by utilizing the spatial and temporalcorrelations in and between the video frames tominimize the embedding distortions. The proposedvideo steganography using IWT achieves zero bit-errorrate between the original and recovered data. Similarly,the reversible data-hiding scheme of Xuang et al.(2005) used IWT and companding technique to embedthe data. With the SMVQ scheme, the receiverreconstructs the original SMVQ compression codes toextract the secret data, resulting in prediction errors.

Elham et al. (2011) utilized a genetic algorithm-based mapping function to embed data in IWTcoefficients in blocks on the cover image. The methodused the genetic algorithm and optimum pixelalgorithm (OPA) to map functions in the dataembedding process in IWT coefficients. The methodachieved better robustness in the frequency domain.Similarly, Jayasudha (2012) presented a steganographyscheme to hide data in IWT coefficients. The methodhas increased hiding capacity and quality by utilizingOPA. However, time frequency is the major constraint.

The previously discussed schemes either affectvideo quality, compromise security, or result in highburden during computation. Thus, improving securityand preserving video quality without affecting thefunctionalities should be further investigated. Recently,many data-hiding schemes have been presented. Themedium used in steganography must be meaningful toensure that hidden data are unnoticeable. Differentfrom the results of data-hiding schemes andconventional data-hiding methods, wavelet transformsare not only random but also significant. Thus, certainmethods based on the transform domain favor the videosteganography approach.

3.5. Markov ModelThe Markov model is based on Markov chains. TheMarkov chain is formed between the inputs and outputswhile extracting the hidden data. However, the Markov

model is simplified by considering all possibilities froma given phase to any other phase to prevent complexityoverhead. Thus, the quality of the text generated bythe Markov chain is modified considerably. Forexample, words such as “the” and “generally” arepotential beginning words of a phrase, but the formershould be more frequent than the latter. This differenceis not preserved by the simplification (Sibastien et al.2003).

Figure 4 shows an example of a Hidden MarkovModel (HMM). The model represents a system thatcan be described at any time based on a set of states.As stated by Lawrence and Rabiner (1989), the HMMcan be represented with N states. Figure 4 shows asample Markov chain with three states, namely, d1,d2, and d3, and the system undergoes a change of stateaccording to a set of probabilities associated with astate. The state changes with a discrete time t. Thesequence of states are denoted as a11, a22, a33, andso on, and the probability of the sequence is the productof the transitions. In this example, the transition statesare a12–a21, a23–a32, a13–a33, and so on.

The method proposed by Lin et al. (2006)maintains the traffic status using the fuzzy C-meanshidden Markov model to support higher levels ofaccuracy for steganography. The proposed algorithmextracts DCT and motion vector features from trafficvideo sequence. The fuzzy C-means hidden Markov

Figure 4: An example of HMM– A Markov chain with threestates (labeled as d1, d2, and d3) with selectedtransition states (Lawrence and Rabiner)

46 � IJCSIE

model is utilized to classify traffic status. Then, adynamic decision process is used to obtain a most likelytraffic state sequence after calculating the videofeatures. The proposed algorithm saved considerablememory resource and provided better computing speed.

Based on the Markov chain source model and DESalgorithm, a text steganography system for spellinglanguages was presented by Weihui (2010). Themethod works reliably, with immunity from regularoperations, such as formatting, compressing, andsometimes manual altering operation in text size, fontcolor, and the space between words. The work issuitable for hiding short information in onlinecommunication, such as email and short messageservice on mobile phones.

Alternatively, hidden Markov models are used todeal with color object tracking (Sibastien et al. 2003).The hidden Markov model handles the data embeddingprocess by tracking the color objects of the carrier.Moreover, the hidden Markov model uses thesemathematical tools to model the object in the spatialdomain. Multidimensional hidden Markov models areadopted to handle multidimensional (color) data.However, steganalysis of the generic LSB-basedcommercial steganographic tools using the Markovmodel is difficult to conduct.

Dekun et al. (2006) presented a steganalysismethod based on the 2D Markov model of thethreshold prediction error image that decodes thesecret data. The empirical transition matrices of theMarkov chain along the horizontal, vertical, anddiagonal directions serve as features for steganalysisin the proposed method. The method obtained a betterdata detection rate.

Another steganalysis scheme by Yun et al. (2007)effectively detected the advanced JPEG steganography.Subsequently, the Markov process is applied to modeldifferences in JPEG 2D arrays for better functioningof the steganalysis task of extracting secret data. Inaddition to the utilization of the differences in JPEG2D arrays, a thresholding technique is employed tominimize the dimensionality of transition probabilityparameters, that is, the dimensionality of featurevectors. Thus, the computational complexity of thescheme is reduced by improving the speed of secretdata retrieval. The survey reveals that the advantageof using the Markov model in video steganography isfast data retrieval. Table 1 summarizes the existingmethods of transform domain based videosteganography.

Video steganography is conducted in the spatialand transform domains. Spatial domain techniques are

Table 1Summary of transform domain based video steganography

Authors Advantages Disadvantages

Mandal, 2012 High embedding capacity, high stability Robustness and lower time frequency

Tamer, 2012 High prediction accuracy High rate of channel bit errors with computationalcomplexity and packet losses

Bin et al., 2011 Reducing embedding distortion and increasing embedding Complex data retrieval if the speed of retrieval is slowefficiency because of database overloads

Yueyun, 2007 Low complexity and low visual distortion Low payload capacity

Aly, 2011 Low prediction error level, noise reduction, and High computational cost and low robustnessquantization error reduction

Hamid et al., High security Image distortion is high, and the addition of noise by the2009 attacker affects the fast data retrieval

Deepika et al., Less distortion and least overhead on the compressed Low clarity of the video sequences and the lossless2012 video size compression quantization is less beneficial

Cheddad et al., Clarity of video files is high; high capacity Hidden data retrieval deals with attacks in the form of2008 additive artificial noise

Qiang and High embedding capacity; detection errors are controlled Low robustness and less securityThomas, 2001

Shen et al., 2012 High efficiency in security of steganography Less payload capacity

Song et al., 2002 High embedding capacity Quality of the carrier is degraded

McKeon, 2007 High security Computational inconvenience

(contd...)

Transform domain based video steganography techniques � 47

Singh and High energy compaction, high security level, and low High computational capacitySiddiqui, 2012 image distortion

Ekta et al., 2010, High-quality image Data embedding capacity of the system is limitedChang et al., 2009

Veerdeep et al., High-quality image Trade-off between quality and embedding capacity2013

Primechaev High embedding capacity In considering multi-image detection, the faster-movinget al., 2007 character results in lower detection accuracy

Hossein et al., Embedded data security is high Low robustness2012

Daniela et al., Security against hacker attack and imperceptible decrease Less security and poor video quality because of the2007 in picture quality occurrence of errors

Linjie et al., Better security High computational complexity2014

Nedal et al., Better embedding rate and minimum distortions Less security and low robustness2009

Xiong et al., Based on the energy consumption and rate distortion High computational complexity2013 model, resource allocation is optimized with limited

energy and bandwidth

Xianhua et al., Better security in the transmission of stego-image visually High computational cost2012 and large payload

Klimis et al., Provides invisibility and resistance against lossy High image distortion2002 transmission or other distortions

Filler et al., Less distortion High computational complexity2011

Pankajakshan High video steganalysis performance and better data Less securityand Ho, 2007 retrieval rate by utilizing temporal correlation

Barve et al., Good performance in securing data Different constraints affect the data hiding process; less2012 security

Souvik and Better security Ineffective data embedding with high computational costGautam, 2010

Hongmei et al., High detection error rate High computational cost2011

Hamsatvan, 2012 Better robustness and imperceptibility Fewer distortions that affect video quality

Anjali et al., High-capacity data-hiding method without sacrificing the Does not protect the image from distortions2014 quality of the cover image

Lakshmi et al., Minimizes the embedding noise in video, supports multiple Time complexity in execution is high2012 resolutions, and has high capacity and security

Elham et al., Better robustness High burden for computation and quality destruction2011

Jayasudha, 2012, High embedding capacity Less security of the data, which affects video qualityAbbas et al.,2006

Sibastien et al., Prevents complexity overhead Less security2003

Lin et al., 2006 High accuracy and better computing speed High computational complexity

Dekun et al., Better data detection rate and lower computational Low robustness2006 complexity of the scheme

Yun et al., 2007 Computational complexity of the scheme is lowered by High computational cost and low robustnessimproving the speed of secret data retrieval

Authors Advantages Disadvantages

(Table 1 contd...)

48 � IJCSIE

sensitive to various image-processing attacks, suchas compression and cropping. Although spatialdomain steganography techniques provide betterhiding capacity, the methods provide less security.Thus, transform domain-based steganographytechniques are well preferred for data hiding.Transform domain-based video steganographytechniques embed the secret data into the transformedcoefficients.

Previous studies show that transform domain videosteganography has a better data detection rate, lowcomputational complexity, and high embeddingcapacity as well as prevents complexity overhead. Themain disadvantages of this method are high timecomplexity in execution, high computational cost, andless security. Although the embedding capacity isstronger, the retrieval speed is slower. Therefore, thisstudy focuses on developing a transform domain-basedvideo steganography system to enhance security, ensurefast data retrieval, enhance time frequencyoptimization, and minimize computational cost.

4. CONCLUSIONIn this paper, a survey of numerous transform domain-based video steganography techniques is provided.Other common steganography mechanisms, such astraditional LSB and Fourier transform, are less robust,and the cover data might be lost through imagemanipulation and simple attacks. These security androbustness limitations in the use of traditional LSB inhiding data necessitate better video steganographymodels with optimized time frequency and highembedding capacity. Based on the investigation, imagesteganography with change in the LSB of each of thepixel intensities of the cover image is useful inembedding data in the cover image. Moreover, DWThas certain benefits. Therefore, incorporating DWTinto the proposed work enhances the imperceptibilityof hidden data to HVS. Furthermore, audiosteganography using DWT in combination with DCTpreserves the correlation among video frames duringpayload extraction and scene change detection. Thus,based on the literature review, DCT can preserve thevideo quality after hiding data and, by integrating withDWT, the method reduces distortions. In addition,DWT in image steganography delivers better resultson image compression, which may also be applied tovideo compression. A secure DWT is significant forcompressed video sequences because it provides highvideo quality without noise. According to the literature

review, the hidden Markov model is capable offacilitating a fast secret data retrieval process.

Transform domain-based video steganographictechniques have three different perspectives, namely,affine transformation, wavelet transforms, and Markovmodels. The importance of various aspects of the threemain models utilized in this study is described in detail.Thus, this study focuses on the key aspects of transformdomain-based video steganography. The ultimateresearch objective is to address the hiding capacity,time frequency optimization, security, quality, dataretrieval rate, and computational complexity.

REFERENCES[1] Abbass A.S., Soleit E.A., and Ghoniemy S.A., 2006. Blind

video data hiding using integer wavelet transforms, UbiquitousComputing and Communication Journal, 2(1), pp.11-25.

[2] Adnan M.A.B and Wafaa M.A.B, 2010. Stego-based-cryptotechnique for high security applications, International Journalof Computer Theory and Engineering, 2(6), pp. 1793-8201

[3] Aly, H.A., 2011. Data hiding in motion vectors of compressedvideo based on their associated prediction error, IEEETransactions on Information Forensics and Security, 6(1), pp.586-590.

[4] Anderson R and Siome G, 2008. Steganography and Steganalysisin Digital Multimedia: Hype or Hallelujah? RITA (Revista deInformática Teórica e Aplicada); 15 (1), pp. 83-110.

[5] Anjali T., Seema R.Y. and Kerhalkar S.G., 2014. Design andanalysis of efficient and robust DWT based imagesteganography technique, International Journal of EmergingTechnology and Advanced Engineering, 4(1), pp.547-552.

[6] Ashok K.B., Srinivas, Niet, Kantepudi, 2012. Steganographybased secrete communication using DWT, InternationalJournal of Engineering Research & Technology, 1(5), pp.1-8.

[7] Barve S., Nagaraj U., and Gulabani R., 2011. Efficient andsecure biometric image steganography using discrete wavelettransform, International Journal of Computer Science andCommunication Networks, 1(1), pp.96-99.

[8] Bin L, Junhui H., Jiwu H., Yun Q.S., 2011. A survey on imagesteganography and steganalysis, Journal of Information Hidingand Multimedia Signal Processing, 2(2).

[9] Chang .C.C., Chen Y.H., Wang Z.H., Li M.C., 2009. Areversible data embedding scheme based on Chinese remaindertheorem for VQ index tables, International Colloquiumon Computing, Communication, Control, and Management, 8-9 Aug. 2009, pp.516 – 518.

[10] Cheddad A., Condell J ; Curran, K. ; McKevitt, P., 2008. Skintone based steganography in video files exploiting the YCbCrcolour space, IEEE International Conference on Multimediaand Expo, pp. 905-908.

[11] Cláudio R.J. , and Jacob S., 2005. Robust watershedsegmentation using wavelet, ACM Journal of Image and visionComputing, 23(7), pp. 661-669.

[12] Daniela S., Mircea S., Bogdan C., Dan C., Razvan C., MihaiM., 2007. Embedding data in video stream using steganography,IEEE-SACI 2007-4th International Symposium on AppliedComputational Intelligence and Informatics, pp. 241-244.

Transform domain based video steganography techniques � 49

[13] Deepika R.C, and Ranjit G, 2012. Data hiding in motion vectorsof compressed video based on their associated prediction error,International Journal of Emerging Technology and AdvancedEngineering, 2(10) p. 586-590.

[14] Dekun Z., Yun Q. S., WeiSu, Guorong X., 2006. Steganalysisbased on Markov model of thresholded prediction-error image,IEEE International Conference on Multimedia and Expo, 2006.

[15] Ekta W, Payal J, Navdeep, 2010. An analysis of LSB and DCTbased steganography, Global Journal of Computer Science andTechnology, 10(1), pp.4-8.

[16] Elham G., Jamshid S., and Nima F., 2011. High capacity imagesteganography using wavelet transform and genetic algorithm,Proceedings of the International Multi Conference ofEngineers and Computer Scientists.

[17] Filler. T, Judas. J, and Fridrich J. 2011. Minimizing additivedistortion in steganography using syndrome-trellis codes, IEEETransactions on Information Forensics and Security, 6(3), pp.920-935.

[18] Hamid. A.J., A.A Zaidan and B.B Zaidanl., 2009. Frameselected approach for hiding data within MPEG video usingbit plane complexity segmentation, Journal of Computing, 1(1).

[19] Hongmei, J, and Yun, 2011. DWT-based video data hidingrobust to MPEG compression and frame loss, InternationalJournal of Image and Graphics, 5(1).

[20] Hossein S., Jafar M., and Mostafa R., 2012. Steganography:DCT coefficient replacement method and compare with JStegalgorithm, International Journal of Computer and ElectricalEngineering, 4(4).

[21] Jayasudha. S., 2012. Integer wavelet transform basedsteganographic method using OPA algorithm, InternationalConference on Computing and Control Engineering 2(4), pp.31-35.

[22] Kamstra, L., Heijmans, H.J.A.M., 2005. Reversible dataembedding into images using wavelet techniques and sorting,IEEE Transactions on Image Processing, 14(12).

[23] Klimis S. N, Nikolaos D. D, Anastasios D. D and Stefanos D. K.,2002. An automatic video-object based steganographic systemfor multi-use message hiding using wavelet transform, IEEEInternational Conference on Systems, Man and Cybernetics.

[24] Lakshmi N., Prabakaran, K., and Bhavani R.G, 2012. A highcapacity video steganography based on integer wavelettransform, Journal of Computer Applications, 5(4).

[25] Lin. C.Y., and Hing Y.T., 2006. A robust image hiding methodusing wavelet technique, Journal of Information Science andEngineering.

[26] Lin L., Jianming H., Qiao H., Jingyan S., 2006. A fuzzy hiddenMarkov model for traffic status classification based on videofeatures, Multiconference on Computational Engineering inSystems Applications.

[27] Linjie G., Jiangqun N., and Yun Q.S., 2014. Uniform embeddingfor efficient JPEG steganography, IEEE Transactions onInformation Forensics and Security, 9(5), p.814-825.

[28] Loay E. G. and Ahmed H. A, 2011. Fast image steganographyusing affine mapping, Diyala Journal for Pure Science.

[29] Mandal J K, 2012. A frequency domain steganography usingZ transform (FDSZT), International Workshop on EmbeddedComputing and Communication System.

[30] McKeon R.T., 2007. Strange Fourier steganography in movies,IEEE International Conference on Electro/InformationTechnology, pp.178-182.

[31] Naoya F., Yoshimichi I., and Noboru B., 2012. Delivery methodfor viewer-specific privacy protected video using discretewavelet transform, 19th IEEE International Conference onImage Processing.

[32] Nedal., Kafri M.S., and Hani Y.S., 2009. Bit-4 of frequencydomain-DCT steganography technique, IEEE, Firs tInternational Conference on Networked Digital Technologies.

[33] Pankajakshan V., and Ho A.T.S., 2007. Improving videosteganalysis using temporal correlation”, Proceedings of IEEE,3rd International Conference on Intelligent InformationHiding and Multimedia Signal Processing, 1, pp. 287-290.

[34] Po Y.C., and Hung J.L., 2006. A DWT based approach forimage steganography, International Journal of Applied Scienceand Engineering.

[35] Primechaev, S. Frolov, A., Simak, B., 2007. Scene ChangeDetection Using DCT Features in Transform Domain VideoIndexing, IEEE EURASIP Conference on Systems, Signals andImage Processing, pp. 369-372.

[36] Qiang C., and Thomas S. H., 2001. An additive approach totransform-domain information hiding and optimum detectionstructure, IEEE Transactions on Multimedia, 3(3).

[37] Ranjeet R, Maninder P and Tarun G, 2013. Zooming DigitalImages using Interpolation Techniques, International Journalof Application or Innovation in Engineering & Management,2(4), pp. 34-45.

[38] Seyedi, S.H., Aghaeinia, H., Sayadian, Abolghasem, 2011. Anew robust image adaptive steganography method in waveletdomain, 19th Iranian Conference on Electrical Engineering.

[39] Shen W., Xianhua S., Xiamu N., 2012. An affine transformationbased image steganography approach, International Journalof Digital Content Technology and its Applications, 6(1).

[40] Sibastien L., Emmanuel B., Thierry B., and Nicole V., 2003. Anew way to use hidden Markov models for object tracking invideo sequences, IEEE International Conference on ImageProcessing, ICIP, Proceedings, 14-17 Sep. Vol.2, p.117-20.

[41] Singh S and Siddiqui T.J, 2012. A security enhanced robuststeganography algorithm for data hiding, International Journalof Computer Science Issues, 9(3), pp.131-139.

[42] Song Y, Cui M and Yao H, 2002. A high capacity data hidingscheme based on DFT. Springer-Verlag, pp. 82-88.

[43] Souvik B., and Gautam S., 2010. Data hiding in images indiscrete wavelet domain using PMM, International Journalof Electrical and Computer Engineering.

[44] Tamer S, 2012. Data hiding in MPEG video files usingmultivariate regression and flexible macroblock ordering, IEEETransactions on Information Forensics and Security, 7(2), p.455-464. (MRFB)

[45] Veerdeep K.M., Harmanjot S.D. , 2013. 32×32 vectorquantization based colour image steganography, InternationalJournal of Enhanced Research in Science Technology &Engineering, 2(9).

[46] Vijay K., and Dinesh K., 2010. Digital image steganographybased on combination of DCT and DWT”, Springer Informationand Communication Technologies Communications inComputer and Information Science, vol. 101.

[47] Weihui Dai, 2010. Text steganography system using Markovchain source model and DES algorithm, Journal of Software,5(7).

[48] Xianhua S., Shen W., XiaMu N., 2012. An Integer DCT andAffine Transformation Based Image Steganography Method,

50 � IJCSIE

IEEE, Eighth International Conference on IntelligentInformation Hiding and Multimedia Signal Processing.

[49] Xiong Z., Fan X., Liu S., Li Y., and Zhang H., 2013. Performanceanalysis for DCT-based coded image communication in wirelessmultimedia sensor networks, International Journal on SmartSensing and Intelligent Systems, 6(1).

[50] Xuang G., Yang. C., Zhen. Y., Shi Y.Q., Ni. Z., 2005. Reversibledata hiding using integer wavelet transform and compandingtechnique, Springer Journal, Digital Watermarking, 3304, p.115-124.

[51] Yueyun S, 2007. A new invertible data hiding in compressedvideos or images, Third International Conference on NaturalComputation (ICNC 2007), vol. 5, pp. 576-580.

[52] Yun Q. S., Chunhua C., and Wen C., A. 2007. Markov processbased approach to effective attacking JPEG steganography,ACM Proceedings of the 8th international conference onInformation hiding.

[53] Zieliñska E, Mazurczyk W, Szczypiorski K, 2014. Trends insteganography, Communications of the ACM, 57(3), pp.86-95.