VLSI Implementation of Visible W atermarking for a Secure ...cessor. Mathai, et. al. [5] present a chip implementation of the same video watermarking algorithm. A DCT domain in-visible

VLSI Implementation of Visible Watermarking for a Secure DigitalStill Camera Design

SarajuP. Mohanty, N. RanganathanandRavi K. NamballaDept.of CSE,Universityof SouthFlorida,Tampa,FL 33620�

smohanty,ranganat,rnamball� @csee.usf.eduAbstract

Watermarking is the processthat embedsdata calleda watermark into a multimedia object for its copyrightprotection. The digital watermarkscan be visible to aviewer on careful inspectionor completelyinvisible andcannotbeeasilyrecoveredwithoutan appropriatedecodingmechanism.Digital image watermarkingis a computation-ally intensivetask and can be speededup sigificantly byimplementingin hardware. In this work, wedescribea newVLSIarchitecture for implementingtwo differentvisiblewa-termarkingschemesfor images.Theproposedhardwarecaninserton-the-flyeitheroneor bothwatermarksinto animagedependingon the application requirement. The proposedcircuit canbeintegratedinto anyexistingdigital still cameraframework. First, separatearchitecturesare derivedfor thetwowatermarkingschemesandthenintegratedinto a unifiedarchitecture. A prototypeCMOSVLSI chip was designedandverifiedimplementingtheproposedarchitecture andre-portedin this paper. To our knowledge, this is thefirst VLSIarchitecturefor implementingvisiblewatermarkingschemes.

1 Introduction

Watermarkingis theprocessthatembedsdatacalledawa-termark,tagor label into a multimediaobjectsuchthatwa-termarkcanbedetectedor extractedlater to make anasser-tion aboutthe object. The objectmay be an image,audio,video,or text. In general,any watermarkingschemeconsistsof threeparts,suchas, the watermark,the encoderandthedecoder. Themarkingalgorithmincorporatesthewatermarkinto the object,whereasthe verificationalgorithmauthenti-catestheobjectdeterminingboththeownerandtheintegrityof theobject.Thewatermarkscanbeappliedeitherin spatialdomainor in frequency domain. According to humanper-ception,thedigital watermarkscanbedividedinto four cate-gories: visible watermark,invisible-robust,invisible-fragileanddual [1, 2]. A visible watermarkis a secondarytranslu-centoverlaidinto theprimaryimageandappearsvisible to avieweroncarefulinspection.

Controller

Interface

and

Watermarking

Controller

Input

Memory

(Flash, SDRAM)

DSP

Processor

Image

Sensors

A/D

Converter

Output

Watermarking Processor

Watermarking

Datapath

Figure1: System Architecture of a Secure Digital StillCamera

Severalsoftwarebasedwatermarkingschemeshave beenpresentedin the literature; however, only a few hardwareschemeshave beenproposed.Strycker, et. al. [4] proposedtheimplementationof a real-timespatialdomainwatermarkembedderanddetectoron a TrimediaTM-1000VLIW pro-cessor. Mathai,et. al. [5] presenta chip implementationofthesamevideowatermarkingalgorithm.A DCT domainin-visible watermarkingchip is presentedby Tsai andLu [6].Garimella, et. al. [7] proposeda VLSI architectureforinvisible-fragilewatermarkingin spatialdomain. Mohanty,et. al. [8] describeda watermarkingchip that hasspatialdomaininvisible robust andfragile watermarkingfunction-alities.

In this work, we focus on the VLSI implementationoftwo spatialdomainvisible watermarkingschemes,onepro-posedby Braudaway, et. al. [9] andthe otherby Mohanty,et. al. [3]. TheVLSI chip caninserteitheroneor both thewatermarksdependingon the requirementof the user. Theproposedwatermarkingchipcanbeintegratedwithin any ex-isting digital still camera.We provide theschematicview ofastill camerathatincludesawatermarkingmodulein Fig. 1,andcall sucha cameraasa ”securedigital still camera”.

2 Watermarking Algorithms

In this section,we outline the watermarkingalgorithmsin brief with the modificationsneededfor hardwareimple-mentation. The notationslisted in Table 1 are neededfordescribingthealgorithms.

Table1: List of Variables used in Explanation�: Original (or host)grayscaleimage�: Watermarkimage(agrayscaleimage)�������: A pixel location��: Watermarkedimage���������: Original imagedimension� ��� : Watermarkimagedimension���: ����� block of theoriginal image�� � : ����� block of thewatermarkimage �� ��: ����� block of thewatermarkedimage��� � : Scalingfactorfor ����� block� �: Embeddingfactorfor ����� block � : Meangrayvalueof theoriginal image � � � : Meangrayvalueof imageblock ���! � � : Varianceof theoriginal imageblock ����#"%$�& : Themaximumvalueof � �� "(' � : Theminimumvalueof � �� "%$�& : Themaximumvalueof � �� "(' � : Theminimumvalueof � ��*) � ' �,+ : Grayvaluecorrespondingto white pixel� � : A globalscalingfactor-/. � -10: Linearregressionco-efficients-(2 � -(3: Linearregressionco-efficients

Algorithm 1 (Proposedin [9]) : The watermarked imageis obtainedby addinga scaledgrayvalueof theimageto thehostimage.Theamountof scalingis donein suchawaythatthealternationof eachoriginal imagepixel occursperceptu-ally by equaldegree. The original formula is simplifed asfollows,wherethescalingfactor 4%5 determinesthestrengthof thewatermark[10].

687:9=@?BADCEFFFFFG FFFFFH6�9I;J=K?BABLNMO9I;J=K?BA1P 5RQTSVUXW,YZK[]\ ^K^V_T`ba 58cedbf gih5jQkSVUeWlY%monp 4%5

for58cedbf gih5jQkSVUeWlYrqtsvu sws�x�xkyiz6�9I;J=K?BABLNMO9I;J=K?BA a 58cedbf gih{K|VZ]\ Z m 4}5

for58cedbf gih5jQkSVUeWlYr~tsvu sws�x�xkyiz

(1)

Theaboveequationis furthersimplifiedto makethehard-wareimplementationeasier. At thesametime, careis takento makesurethatthehardwareis asaccurateasthesoftwareimplementations.We assume

6�XIC y�y andsimplify theaboveequationsto thefollowing.

687:9I;J=K?BADC EFFG FFH6�9I;J=K?BABLPw^]\ |V{V_V^ ` MO9=@?BA�9�6�9I;J=K?BA@A np

for69=@?BA q u yixw6�9I;J=K?BABLP�{K|VZ]\ Z ` MO9=@?BA@6�9I;J=K?BA

for69=@?BA ~ u yixw (2)

We further simplify the above expressionsand remove thecubic root function with a piecewise linear model. Wedivide the gray values range s =T6 eI* to four ranges,

suchas s = 5jQTSKUeWlY , 5jQTSVUXW,Y = 5jQkSVUeWlY , 5jQTSKUeWlY = Z 5jQTSVUXW,Y , and Z 5jQkSVUXW,Y =T6 8eI . We fit four linear regressionco-efficientsthatbestapproximatesthecubicroot in eachof theseranges.Moreover, we roundupthefractioninvolvedin thecompari-sonoperationandthe final simplified expressionthat is im-plementedusinghardwareis asfollows.

6 7 9I;J=K?BADC

EFFFFFFFFFFFFFFG FFFFFFFFFFFFFFH

6�9I;J=K?BABLPJ�{K|VZ]\ Z ` MO9=@?BA169=@?BAfor

69=@?BA ~ 6�9I;J=K?BABL P K^]\ |V{V_K^T` MO9I;J=K?BA16�9I;J=K?BAfor

t6�9I;J=K?BA ~z�6�9I;J=K?BABL P w n^]\ |V{V_K^T` MO9I;J=K?BA16�9I;J=K?BAfor z� 69=@?BA ~  x6�9I;J=K?BABLP(w p^]\ |V{V_K^ ` MO9I;J=K?BA16�9I;J=K?BAfor   x 69=@?BA ~¡ 8¢ 6�9I;J=K?BABL P( K�£^]\ |V{V_K^T` MO9I;J=K?BA16�9I;J=K?BAfor  ¢ ¤69=@?BAo y�z

(3)

Algorithm 2 (Proposedin [3]) : Thepixel grayvaluesaremodifiedbasedon the local andglobal statistics. The wa-termakinginsertionprocessconsistsof the following steps.Both thehostimage(oneto bewatermarked)

6andthewa-

termark(image)M

aredividedinto blocksof equalsizes(thetwo imagesmay be of unequalsize). Let ¥§¦ denotethe ¨ Iblock of the original image

6and ©�¦ denotethe ¨ I block

of thewatermarkM

. For eachblock ( ¥§¦ ), thelocalstatistics;meanªB¦ 5 andvariance«�¦ 5 arecomputed.Theimagemeangray value ª 5 is also found out. The watermarked imageblock is obtainedby modifying ¥ ¦ asfollows.

¥ 7 ¦ C 4(¦¥§¦ L¬ ¦©�¦ ¨ C   =* ulueu (4)Where,4 ¦ and ¬ ¦ arescalingandembeddingfactorsrespec-tively, dependingon ª ¦ 5 and « ¦ 5 of eachhostimageblock.

The choiceof 4 ¦ and ¬ ¦ are governedby certainchar-acteristicsof humanvisualsystem(HVS) andmathematicalmodelsareproposedsothattheperceptualqualityof theim-agearenotdegradeddueto watermarkaddition.The 4(¦ and¬ ¦ areobtainedasfollows. The 4%¦ and ¬ ¦ for edgeblocksaretakento be 4 d®�¯ and ¬ d g respectively. The 4(¦ and ¬ ¦arefoundoutusingthefollowing equations.

4 ¦ C °±²´³ kµ¶�· PK¸ 9/¹ª ¦ 5 ¸ ¹ª5 A `¬ ¦ C ¹«¦ 5 P   ¸ µ¶w· P@¸ 9º¹ªB¦ 5 ¸ ¹ª 5 A `` (5)Where,

¹ª ¦ 5 and ¹ª5 are normalisedvaluesof ª ¦ 5 and ª5 ,and

¹« ¦ 5 arenormalisedlogarithmvaluesof « ¦ 5 . The 4 ¦ and¬ ¦ arescaledto the ranges( 4}d g , 4%d®�¯ ) and( ¬ d g , ¬ d®�¯ )respectively, where 4}d g and 4%d®�¯ areminimumandmax-imum valuesof scalingfactor, and

¬ d g and ¬ d®�¯ aremin-imum andmaximumvaluesof embeddingfactor. Thesepa-rametersdeterminetheextentof watermarkinsertion.A lin-eartransformationis usedto scalecurrent 4 ¦ and ¬ ¦ valuesto theranges( 4 d g , 4 d®�¯ ) and( ¬ d g , ¬ d®�¯ ), respectively.

Let currentvaluesof 4 ¦ be written as 4%»¦ , and 4%»d g and4(»do�¯ , respectively denotethe currentminimum andmaxi-mum values. Similarly, let currentvaluesof

¬ ¦ be writtenas¬ »¦ , and ¬ »d g and ¬ »d®�¯ , respectively denotethe current

minimum andmaximumvalues. The 4 ¦ and ¬ ¦ valuesarescaledasfollows.4 ¦ C a �¼}½@¾´¿�¼ UXÀwÁ¼}½@¾ ¿ Á¼ UXÀ m 4%»¦ L a 4%d®�¯ ¸ a �¼B½K¾Â¿�¼ UXÀwÁ¼B½K¾ ¿ Á¼ UXÀ m 4%»d®�¯ m¬ ¦ C avà ¼B½K¾ ¿ à ¼ UeÀà Á¼B½K¾ ¿ à Á¼ UeÀ m ¬ »¦ L a ¬ d®�¯ ¸ aÄà ¼}½@¾ ¿ à ¼ UXÀà Á¼}½@¾ ¿ à Á¼ UXÀ m ¬ »d®�¯ m (6)

We usedfirst-orderderivatives for edgedetection. Forhorizontaledgedetection,we computethehorizontalgradi-ent as: Å v9I;J=K?BAÆCÇ69=@?BA ¸ 6�9I;ÈL   =@?BA . The verticalgradientis computedas ÅÊÉ 9=@?BADCË6�9I;J=K?BA ¸ 69=@?¤L   Afor verticaledgedetection.Theamplitudeof anedgeis cal-culatedas, Å 9I;J=K?BA/CÌ Å �9=@?BA8Ì8LÌ ÅÊÉ 9=@?BA]Ì . Themeanamplitudefor a block is computedas,ÅÊÍ C °Î%Ï1ÐÄÎ%ÏÒÑ d Ñ g Å 9=@?BA (7)Whenthe meanamplitudefor a block exceedsa predefinedthreshold,we declareit asan edgeblock. The valuesof

;and

?correspondto thepixel locationsof individual blocks

with referenceto theoriginal imagepixel location.Themeangrayvalueof a block is calculatedastheaver-

ageof the grayvaluesof all pixels in the imageblock. Themeangrayvaluesarenormalizedwith purewhite pixel grayvalue.Thus,thenormalizedmeangrayvaluesof ablock is,¹ª ¦ 5 C °Î Ï ÐÓÎ Ï a °5jQkSVUeWlYwm Ñ d Ñ g 6�9I;J=K?BA (8)Where,

;and

?are the pixel locationsof the ¨ I image

block,sameastheir locationsin theoriginal image.Thenor-malizedstandarddeviation of grayvaluesfor the ¨ I blockis calculatedasfollows.¹« ¦ 5 C °Î Ï ÐÓÎ Ï a 5jQTSVUXW,Ywm Ñ d Ñ gÔÔ 6�9I;J=K?BA ¸ 5RQTSVUXW,Y ÔÔ (9)Theexponentialtermin Eqn. 5 is approximatedasa Taylorseriesuptothesquareterm.

In stepthreeof theinsertionalgorithm,scalingneedsto bedoneusinga linear transformation,to find thecurrentmini-mum andmaximumvaluesfor both 4%¦ and ¬ ¦ over all theblocks. The hardwareperformanceis going to be severelydegradedsinceit hasto wait till all thepixelsof the imagesareprocessedto derive the local statisticsof all the blocks.So, we modify the above Eqn. 5 to ensurethat the perfor-manceof thehardwareis improvedwith no compromiseonthequality. Wefind 4 ¦ and ¬ ¦ usingthefollowingequations.4(¦ C 4 d g LJ9 4 d®�¯ ¸ 4 d g A °±²³ µ¶w· P@¸ 9/¹ªB¦ 5 ¸ ¹ª 5 A `¬ ¦ C¬ d g L9I¬ d®�¯ ¸ ¬ d g A¹« ¦ 5 P   ¸ µ¶�· P ¸ 9Õ¹ª ¦ 5 ¸ ¹ª5 A `` (10)Extensive simulationsfor variousimagesshow that the 4 ¦and

¬ ¦ obtainedusingEqn. 6 andEqn. 10 arecomparable(maximumdifferenceis ykÖ [1]).

Comparator

Register

FileMultiplier Multiplier

Multiplier

Adder

WI (m,n)

αI

I(m,n) W(m,n)

(a) For Algorithm 1

αk β k

Edge Detection

Unit

0 1 0 1

minβmaxα

Multiplier Multiplier

Adder

WI (m,n)

αk

β kand CalculationUnit

I(m,n) W(m,n)

(b) For Algorithm 2

Figure2: Datapath Architectures for the Algorithms

3 VLSI Architecture

We developan architecturefor the first algorithmshownin Fig. 2(a). A register file is usedto storethe constantsneededto scaletheimage-watermarkproductin Eqn. 3. Westorethe constants

°{K|VZ]\ Z , ^�\ |*{V_K^ , n^]\ |V{V_V^ , p^�\ |*{K_V^ , and £^�\ |*{V_K^and the other constant4%5 is assumedasa parameter. Thecomparatoris usedto determinetherangein which a partic-ular pixel gray valuelies, suchthat an appropriateconstantcanbe picked up from the registerfile. The left sidemul-tiplier calculatesappropriateconstanttimes the host imagepixel grayvaluesandtheright sidemultiplier is usedto find4 5 timesthewatermarkimagepixel grayvalue. Theresultsof the above two multipliers are fed to the third multiplierwhicheffectively calculatestheproductof constants,4%5 , thehostimagepixel grayvalue,andthewatermarkimagepixelgrayvalue,respectively. Theproductis addedto thehostim-agepixel grayvaluesusingtheadderto obtainwatermarkedimagepixel grayvalues.Theabovedescribedprocessis car-ried out for all thepixelsto obtainthewatermarkedimage.

Thearchitecturebeingproposedfor thesecondalgorithmis shown in Fig. 2(b) which presentthe operationat pixellevel. The ” 4%¦ and ¬ ¦ calculationunit” computesthe 4(¦and

¬ ¦ valuesfor the ¨ I non-edgeblock usingexpressionin Eqn. 10. The”edgedetectionunit” determinesif a blockis an edgeblock or non-edgeblock. If the Å Í exceedsanuserdefinedthreshold,thenit is an edge-block.Larger thethresholdmorearetheblocksdeclaredasedge-blocks.Themultiplexorshelpin selectingthescalingandembeddingfac-tors betweenthe edgeandnon-edgeblocks. The multiplieron theleft calculatesthescalingfactorstimesthehostimagepixel grayvalue.Theright sidemultiplier multipliestheem-beddingfactorwith the watermarkimagepixel gray value.Theproductsfrom thesetwo multipliersareaddedusinganadderto find thewatermarked imagepixel grayvalue. Thisprocessis repeatedfor all pixelsin ablock,andsubsequentlyfor all theblocksin theimage.

Adder / Subtractor

Adder / Subtractor

Adder

Accumulator

Multiplier

kIµ

<

( − 0.5 )

kIµ

<

Multiplier

16384

1

Adder / Subtractor

Adder

Accumulator

1

8192

kIσ

<

Multiplier

Multiplier

αmax

αmin( − )

Multiplierβ min

Adder

αmin

Adder

β max β min( − )

β k αk

I(m,n)

0.5

1

Divider

128

Exponential Unit

(a) × � and Ø � Calculation

Adder / SubtractorAdder / Subtractor

Adder

Adder

Accumulator

Multiplier

1

64

Comparator

ThresholdAmplitude

µG

I(m+1,n) I(m,n) I(m,n+1)

G(m,n)

Edge or Non−edge Block

(b) EdgeDetection

Figure3: Individual Datapath Units for Algorithm 2

4 ¦ and ¬ ¦ calculationunit: The architecturaldetails ofthe” 4 ¦ and ¬ ¦ calculationunit” areshown in Fig. 3(a).Thehardware implementsthe calculationfor 4 ¦ and ¬ ¦ repre-sentedas Eqn. 10 for one block at a time. The left sideadder-accumulatorcombinationfinds the sumof all the im-agepixel gray valuesfor a block. After the sum is multi-

plied with a °Î Ï ÐÓÎ ÏJÙ °5 QkSVUeWlYwm , we get thenormalisedmeangrayvalueof ¨ I block denotedby ¹ªB¦ 5 . Sincewe have as-sumedblock size of xÚNx , and 6 8eI as y�z , this evalu-atesto

°° ^KZV[ . It may be notedthat 6 8eI is ywy , but using y�z makeshardwareimplementationeasier, the latterbeingrepresentableasa power of two. In the original algorithm9/¹ª ¦ 5 ¸ ¹ª5 A is thedeviation of a meangrayvalueof a blockfrom theimagemeangrayvalue.Weareevaluatingthedevi-ationof meanblock grayvaluefrom mid-intensityof 5RQTSVUXW,Yfor simplicity, . Thus,

9/¹ª ¦ 5 ¸ ¹ª5 A is computedas 9º¹ª ¦ 5 ¸ sÄuÛy A ,whennormalisedwith

6]8eI. This assumptionaccelerates

the hardwareperformanceto a greatextentsincethe block-by-blockwatermarkingcanbeperformedwithoutwaitingforthe global imagestatisticscomputedover the whole imagebeforethe watermarkinsertioncanbe performed. The ex-pressionµ¶�· P ¸ 9Õ¹ªB¦ 5 ¸ ¹ª 5 A ` is computedusingthe”expo-nentialunit”.

Theadder/subtractorunit findstheimagepixel grayvalueabsolutedeviation from 5jQTSVUXW,Y . The adder-accumulatorfol-lowing this accumulatethe Ñ d Ñ g ÔÔ 69=@?BA ¸ 5jQkSVUeWlY ÔÔ fora block. When this sum is multiplied with a °Î Ï ÐÓÎ Ï m Ùa 5jQkSVUeWlYwm , which is xÄ ¢ for our case,we getthenormalisedstandarddeviation

¹«�¦ 5 . Theright sidedivider dividesexpo-

nentialvaluecomputedbeforeby¹« ¦ 5 . Thequotientis then

multiplied with 4%do�¯ ¸ 4%d g . Theabove productis addedto 4}d g to evaluate4 ¦ expressedin Eqn. 10. Theexponen-tial unit resultis fed to a adder/subtractoron left sidewhichfindsits differencefrom 1. Theresultis thenmultipliedwith¹« ¦ 5 obtainedfrom thecomputationsperformedbefore.Theproductobtainedis thenmultipliedwith

¬ d®�¯ ¸ ¬ d g . Thisproduct is then addedto

¬ d g which in turn gives the re-quired

¬ ¦ asperEqn.10.Edgedetectionunit: Thearchitecturefor determiningif a

blockis anedgeor non-edgeblockis shown in Fig. 3(b). Theabsolutevaluesof thehorizontalgradient

Ì Å 9=@?BA8Ì andtheabsolutevalueof verticalgradient

Ì Å É 9I;J=K?BA]Ì arecalculated.The amplitudeof an edge Å 9I;J=K?BA is calculatedusing thefirst adder. Then, the adder-accumulatorcombinationfindsthesumof Å 9=@?BA for all pixelsof a block. Theabovesumwhenmultiplied with a °Î Ï ÐÄÎ Ï m 9jC zi A , we get the meanamplitudeÅÊÍ for a block. Thecomparatorcomparesthe ÅÊÍvalueswith anuserdefinedthresholdanddeclarestheblockasanedgeor non-edgeblock.

The individual datapathsfor both the algorithms arestitchedtogetherusing multiplexors and a combineddata-path shown in Fig. 4(a) is obtained. Both the datapathssharethe samemultipliers, as it is evident from Fig. 4(a),the multiplexors help in selectinginput for the multipliers(WhenSelectis ”0” algorithm1 is used).Thecontrollerthatdrivesthedatapathis shown in Fig. 4(b). Thecontrollerhassix states,suchasInit, ReadBlock,WriteBlock, ReadPixel,WritePixel, andDisplayImage.WhentheStartsignalis ”1”thewatermarkingprocessis initiated. Dependingon theSe-lect signaloneof the watermarkingschemesis chosenandthe correspondingdatapathneedsto be driven to carry outthewatermarkingprocess.

4 Prototype Chip Implementation

The implementationof the watermarkingdatapathandcontrollerwascarriedout in the physicaldomainusingtheCadenceVirtuosolayouttool usingbottom-to-tophierarchi-caldesignapproach.Thedesigninvolvedtheconstructionoffour mainunits,suchastheexponentialunit, theedgedetec-tion unit, the 4(¦ and ¬ ¦ calculationunit, registerfile, andtheaccumulator. All of theaboveunitshavemultipliers,adders,adder/subtractor, divider, comparator, andsoon. Thesesmallfunctionalunitsarelaid outindividually throughmodulariza-tion andlaterinterfacedwith eachotherto getthefour abovementionedunits. The datapathand the controller are con-structedusingthe main units andthe functionalunits. Thelayoutsof thegatesat thelowestlevel of hierarchyaredrawnusingtheCMOSstandardcell designapproach.Wedesignedour own standardcell library containingbasicgates,suchasAND, OR, NOT. The completelayout of the watermarking

αk β k

Edge Detection

Unit

0 1 0 1

minβmaxα

αk

β kand CalculationUnit

0 1

Register File

Comparator

0 1

αI

Multiplier Multiplier

Multiplier

0 10 1

I(m,n) W(m,n)Select

Adder

WI (m,n)

(a) Datapath for Algo-rithms1 and2

Read

Pixel

Read

Block

Write

Block

Display

Image

Write

Pixel

Init

BlockCompleted=1

BlockCompleted=0

BlockCompleted=1

ImageCompleted=1

BlockCompleted=1

ImageCompleted=1

ImageCompleted=1

ImageCompleted=0

ImageCompleted=0

BlockCompleted=0

Start=0

Select=1

Start=1Start=1

Select=0

ImageCompleted=0

(b) Controllerfor MergedDatapath

Figure4: Architecture for the Proposed Processor

chip is givenin Fig. 5(a)andthefloor planof thechip is pro-videdin Fig. 5(b). Table2 showstheoveralldesigndetailsofthechip andthecorrespondingpin diagramis shown in Fig.6. Thechip statisticsareobtainedusingHSPICEfor sÄu wy ªMOSISSCN3MSCMOStechnology.

Thedatapathconstructioninvolvestheimplementationoftheproposedarchitecturein theprevioussection.Thefunc-tionalunitsare8-bit ripplecarryadders,8-bit multipliersand8-bit adder/subtractor. Theadder/subtractorunit is obtainedfrom theadderusingXOR gates[11]. An 8-bit parallelarraymultiplier is obtainedfrom full-addersandAND gatesto im-plementmultiplication operationswith reduceddelay [12].Thedivider is implementedusingtheshift andsubtractlogicfor thedivision [11]. Thecomparatorwasdesignedto com-parevaluesof two 8-bit numbersfor greater-than,equalto,or less-thanrelations.First, a single-bitcomparatorwasde-signedto comparethevaluesof two single-bitnumbers,andlater, instancesof thismodulewerecascadedto comparetwo8-bit numbers,startingfrom themost-significantbit positionandproceedingtowardstheleast-significantbit position.

Theaccumulatoris implementedasa14-bit registerto ac-

(a)Chip Layout

αk

andβk

CalculationUnit

Edge−Detection

Unit

Other ComponentsController

(b) FloorPlan

Figure5: Proposed Watermarking Chip

Table2: Overall Statistics of the Watermarking Chip

Area Ü�Ý Ü]Þ �ß Ý à8á �� 0Numberof gates

ß à]Þ´â8áClock frequency

ß á ß Ý ßã8äNåçæNumberof I/O pins

ãßPower â�Ý á ß àâ � �

commodatea maximumvalueof zièÚ yiz . The maximumvalueoccurswheneachpixel in a xÒÚJx block assumesthevalueof purewhite pixel grayvalue. The registerfile is anaddressablearrayof 8-bit registers(words) [12]. Basedonthe addressspecifiedand a Read/Writeselectline, at anytime, a valuecanbe eitherwritten to or readfrom the reg-ister file. Here,we useda 5-word registerfile to storethefive differentconstants,suchas

°{K|VZ]\ Z , ^]\ |V{V_K^ , n^�\ |*{V_K^ , p^�\ |*{K_V^ ,and £^]\ |V{V_V^ , in Eqn. 3. Multiplexorsareusedat appropriateplacesin thedesignto selectoneof theincominglines.Eachof suchmultiplexor is implementedusinga combinationoftransmissiongates. Threeasynchronouslyresettableregis-tersaredesignedto encodethe five statesof the controllerdepictedin Fig. 4(b). The threeregisterscouldbe resetbytheuserto returnthecontrollerto its intial stateat any timeandfrom there,the watermarkingfunction could be startedafresh.

5 Results and Conclusions

Thefunctionalunitsaresimulatedindividuallybeforetheyareintegratedto developthewholechip. Thefunctionalveri-ficationof thewholechip is doneby performingwatermark-ing on varioustest images. The test imagesare borrowed

Second / First

αminαmaxβ minβ max

αI

DataOut

Visible

{ImageDataInWatermarkDataIn

Start

Reset

Clock

Spatial Domain

Watermarking

Chip

Busy

DataReady

Figure6: Pin diagram for the Proposed Chip

(a)OriginalLena (b) WatermarkImage

(c) UsingAlgorithm 1 (d) UsingAlgorithm 2

Figure7: Original and Watermarked Images

from [3, 1] andof dimension y�zéÚ yiz . Fig. 7 shows one

test image,a watermarkimageused,andwatermarked im-ages.The watermarked imageis alsoshown in Fig. 7. Forfirst algorithm,the valuesof 4 d g , 4 d®�¯ , ¬ d g , and ¬ d®�¯areassumedas svu ¢ky = sÄu ¢�x = sÄu s , and sÄu skê , respectively, andfor secondalgorithm 4 5 is svu sw . The visual inspectionofthewatermarkedimagesprovesthatwatermarkingis abletopreservethequalityof theimagewhile explicitly proving theownership. Of the variousquantitative measuresavailableto quantify thequality of the watermarked images,we usedsignal-to-noiseratio

9Rë1ìîíÊAassuggestedby [5, 3, 1]. We

calculatedtheë1ìîí

usingtheoriginal andthewatermarkedimageusinga softwaresimulator. Simulationresultsshowthatthe

ë1ìífor variouswatermarkedimagesis in therange

of swïwð to y�ïwð .

In this paper, we havepresenteda watermarkingchip thatcanbeintegratedwithin a digital cameraframework for wa-termarkingimages.Thechip hastwo differenttypesof wa-termarkingcapabilities,both in spatialdomain. Out of thetwo watermarkingschemesimplemented,the first onedoespixel-by-pixel processingandthesecondoneis a block-by-block processingalgorithm. Additional work needsto bedoneto developblock-by-blockoperationfor the first algo-rithm so that high performancehardwarecan be designed.However, boththealgorithmsarecomparablefrom the

ë1ìîípointof view.

References[1] S. P. Mohanty, “Watermarkingof Digital Images,” M.S. the-

sis,IndianInstituteof Science,Bangalore,India,1999.

[2] N. MemonandP. W. Wong, “ProtectingDigital MediaCon-tent,” Comm.of theACM, vol. 41,no.7, pp.34-43,Jul 1998.

[3] S. P. Mohanty, K. R. Ramakrishnan,andM. S. Kankanhalli,“A DualWatermarkingTechniquefor Images,” in Proc.of the7thACM Intl. MM Conf. (Vol. 2), 1999,pp.49-51.

[4] L. D. Strycker, et.al., “Implementationof aReal-TimeDigitalWatermarkingProcessfor BroadcastMonitoring on TrimediaVLIW Processor,” IEE Proc. on Vision, Image and SignalProcessing, vol. 147,no.4, pp.371-376,Aug 2000.

[5] N. J. Mathai,D. Kundur, andA. Sheikholeslami,“HardwareImplementationPerspectivesof Digital Video WatermarkingAlgortithms,” IEEETrans.onSignalProcessing, 2003.

[6] T. H. TsaiandC. Y Lu, “A SystemLevel Designfor Embed-dedWatermarkTechniqueusingDSC System,” in Proc. oftheIEEE Intl. Workshopon IntelligentSignalProcessingandCommunicationSystem, 2001.

[7] A. Garimella,et.al., “VLSI Impementationof OnlineDigitalWatermarkingTechniquesWith DifferenceEncodingfor the8-bit GrayScaleImages,” in Proc.of theIntl. Conf. on VLSIDesign, 2003,pp.792-796.

[8] S. P. Mohanty, N. Ranganathan,andR. K. Namballa, “VLSIImplementationof InvisibleDigital WatermarkingAlgorithmsTowardsthe Developementof a SecureJPEGEncoder,” inProc. of the IEEE Workshopon Signal ProcessingSystems,2003,pp.183-188.

[9] G. W. Braudaway, K. A. Magerlein,and F. Mintzer, “Pro-tectingPublicly Available Imageswith a Visible ImageWa-termark,” in Proc.of theSPIEConf. on Optical SecurityandCounterfietDeterrenceTechnique, 1996,pp.126-132.

[10] J. MengandS. F. Chang, “EmbeddingVisible VideoWater-marksin theCompressedDomain,” in Proc.of theIntl. Conf.on Image Processing(Vol. 1), 1998,pp.474-477.

[11] V. P. Nelson,et.al., Digial Logic AnalysisandDesign, Pren-tice Hall, UpperSaddleRiver, New Jersey, USA, 1995.

[12] N. H. E. WesteandK. Eshraghian,Principlesof CMOSVLSIDesign: A SystemsPerspective, Addison Wesley, Boston,MA, USA, 1999.