A ROBUST DIGITAL IMAGE COPYRIGHT PROTECTION USING 4-LEVEL DWT ALGORITHM

Transcription

1 Inernaional Journal of Advanced Technology in Engineering and Science wwwiaescom Volume No, Issue No, November 4 ISSN online: A ROBUST DIGITAL IMAGE COPYRIGHT PROTECTION USING 4-LEVEL DWT ALGORITHM Y Jaya lakshmi, Priscilla Dinkar Moyya, TV Madhusudhana Rao 3 ABSTRACT PG Suden [DECS] Dep of ECE, Assisan professordep of ECE, Associae professordep of CSE 3 Thandrapaparaya Insiue of science & echnology, Bobbili, AP, India This paper presens a Digial Image copyrigh proecion using 4-Level DWT for ownership auhenicaion LL 4 sub-band is used for inser he informaion in he original image using modulus hiding Algorihm Original image or hidden informaion is no necessary for exracing he binary waermark Proposed algorihm is robus o common image processing aacks like, median filering and Gaussian filering Experimenal resuls show ha here is a significan improvemen wih he proposed algorihm over he exising schemes in erms of Normalized Correlaion Coefficien NCC and Peak Signal o Noise Raio PSNR Keywords -- Digial Copyrigh Proecion Informaion Hiding Process, Discree Wavele Transform, Haarwavele, Modulus Hiding, Normalized Correlaion Coefficien Ncc, Peak To Signal Raio Psnr I INTRODUCTION The rapid growh of he Inerne and digial media manifess iself in widespread public forms such as he digial image, he MPEG, and so on, because digial media are easy o copy and ransmi Many researchers are aware of he issues of copyrigh proecion, image auhenicaion, proof of ownership, ec Hence, here are many soluions ha have been proposed The copyrigh proecion echnique is one of he soluions This echnique embeds informaion so ha i is no easily percepible; ha is, he viewer canno see any informaion embedded in he conens There are several imporan issues in he copyrigh proecion sysem Firs, he embedded waermark should no degrade he qualiy of he image and should be percepually invisible o mainain is proecive secrecy Second, he waermark mus be robus enough o resis common image processing aacks and no be easily removable; only he owner of he image ough o be able o exrac he waermark Third, he blindness is necessary if i is difficul for us o obain he original image and waermark A copyrigh proecion echnique is referred o as blind if he original image and waermark are no needed during recovery The process of digial copyrigh proecion involves he modificaion of he original mulimedia daa o embed a waermark conaining key informaion such as auhenicaion or copyrigh codes The hiding mehod mus leave he original daa percepually unchanged, ye should impose modificaions which can be deeced by using an appropriae recovery algorihm Common ypes of signals o waermark are images, music clips and digial video In his work, applicaion of he digial copyrigh proecion o sill images is considered The maor echnical challenge is o design a highly robus digial copyrigh proecion echnique, which discourages copyrigh infringemen by making he process of copyrigh proecion removal edious and cosly [] Curren echniques described in he lieraure for he copyrigh proecion of images can be grouped ino wo classes: 75 P a g e

2 Inernaional Journal of Advanced Technology in Engineering and Science wwwiaescom Volume No, Issue No, November 4 ISSN online: ransform domain mehods [, 3] which embed he daa by modulaing he ransform domain coefficiens and spaial domain echniques [, 4] which embed he daa by direcly modifying he pixel values of he original image To achieve robusness, daa should be embedded in he regions ha conain some imporan characerisics informaion of he cover image [5] This is based on he basic assumpion ha so long as he imporan characerisics of he differen regions in he cover image are no drasically changed, hidden daa can be exraced faihfully Some of he maor characerisics of a gray level image are edge, exure and high gray level curvaure poins, ec [6] Robusness is furher improved if he waermark informaion is embedded in he suiable ransform coefficiens of he image characerisics Several digial image copyrigh proecion schemes using discree Fourier ransform DFT [7], Fourier Mellin [8], discree cosine ransform DCT [9, ], and discree wavele ransform DWT [, ] have already been repored in he lieraure In his work, a robus copyrigh proecion algorihm based on DWT is presened o embed a 3 x 3 binary waermark ino a 5 x 5 gray scale hos image using he modulus hiding scheme Fourh level approximaion sub band LL 4 is seleced for hiding he waermark Each coefficien in he seleced sub band is modified according o he waermark bi during hiding process A modulus decoder is used o exrac he waermark bis from possibly aacked image Experimenal resuls show ha he proposed mehod decreases he disorion of he hos image and is effecively robus agains compression, average filering, Gaussian filering and he PSNR value of he waermarked image is greaer han 4dB The res of he paper is organized as follows: In secion, a brief background maerial abou DWT is presened In secion 3, informaion hiding and recovery is described Experimenal resuls are presened in secion 4 o demonsrae he much improved performance of he proposed mehod in comparison wih he exising echniques Robusness agains he mos common aacks is also presened Finally, he conclusions are given in secion5 II DISCRETE WAVELET TRANSFORM Waveles are special funcions which, in a form analogous o sines and cosines in Fourier analysis, are used as basal funcions for represening signals [] For -D images, applying DWT corresponds o processing he image by -D filers in each dimension The filers divide he inpu image ino four non-overlapping muliresoluion sub-bands LL, LH, HL and HH The sub-band LL represens he coarse-scale DWT coefficiens while he sub-bands LH, HL and HH represen he fine-scale of DWT coefficiens To obain he nex coarser scale of wavele coefficiens, he sub-band LL is furher processed unil some final scale N is reached When N is reached we will have 3N+ sub-bands consising of he muli-resoluion sub-bands LL N and LH x, HL x and HH x where x ranges from unil N Due o is excellen spaial-frequency localizaion properies, he DWT is very suiable o idenify he areas in he hos image where a waermark can be embedded effecively In paricular, his propery allows he exploiaion of he masking effec of he human visual sysem such ha if a DWT coefficien is modified, only he region corresponding o ha coefficien will be modified In general mos of he image energy is concenraed a he lower frequency sub-band LL N and herefore hiding waermarks in ha sub-band may degrade he image significanly Hiding in he low frequency sub-band, however, could increase robusness significanly On he oher hand, he high frequency sub-bands HH x include he edges and exures of he image 76 P a g e

3 Inernaional Journal of Advanced Technology in Engineering and Science wwwiaescom Volume No, Issue No, November 4 ISSN online: P a g e and he human eye is no generally sensiive o changes in such sub-bands This allows he waermark o be embedded wihou being perceived by he human eye The compromise adoped by many DWT-based copyrigh proecion algorihms, is o embed he waermark in he middle frequency sub-bands LHx and HLx where accepable performance of impercepibiliy and robusness could be achieved [3, 4, 5] HAAR WAVELET TRANSFORMATION: Waveles are a special kind of funcions which exhibis oscillaory behavior for a shor period of ime and hen dissappear The oldes and mos basic wavele sysem is named Haar wavele ha is a group of square waves wih magniude of ± in he inerval [, [, ] Haar wavele has Haar wo funcions namely Haar scaling equaion and wavele funcion, hey are represened as,,, for oherwise,,, for oherwise for All he oher subsequen funcions are generaed from k 3 Where, k i And k All he Haar waveles are orhogonal o each oher From he Haar funcions, he scale equaion and wavele equaions are obained as follows From he equaions 4 and 5, we ge low pass componens as h h and high pass componens as, g g Using hese filer componens Haar wavele decomposiion over one-dimension digial signals can be expressed as, T C D C And 4 N N T

4 Inernaional Journal of Advanced Technology in Engineering and Science wwwiaescom Volume No, Issue No, November 4 ISSN online: From he equaion 6, we can compue he approximaion coefficiens vecor C and deail coefficiens vecor D, hrough operaing Haar wavele decomposiion over he vecor C+ III INFORMATION HIDING AND RECOVERY ALGORITHM a A hos image is specified by an m X n marix This marix is he original image o be waermarked The digial informaion hiding process is divided ino hree seps and is briefly described below Sep Discree -D Wavele Transform: The single level discree periodic -D wavele ransform is applied o he original image and he coefficien marices are sored This operaion is reapplied o he resuling LL band for hree more ieraions, producing a four-level DWT wih several coefficien marices Figure shows he layou of he resuling coefficien marices Sep Embed he firs waermark in he LL 4 band: The waermark, denoed by X, is embedded ino he LL 4 band, denoed by C, according o developed by Kang e al [], where Ci, and C'i, denoe he ampliude of he i, h elemen in C and C', respecively, C' being he waermarked LL 4 band; α is a parameer relaed o he waermark embedding srengh; and xi, denoes he i,h elemen in X Noe ha C and X mus have he same dimensions Sep 3 Perform inverse DWT o obain he waermarked image: The disinc sub bands of he image are broken up and he inverse DWT is performed in a reverse manner, o recompose he LL 4, LH 4, HL 4, and HH 4 bands firs, hen hose of he hird level, he second level, and finally he lowes level ransform, back o achieve a waermarked image b The recovery is blind wih respec o he original image Recovery is compleed wih wo seps, as follows: Sep Perform DWT on he waermarked image: The waermarked image is ransformed ino wavele sub bands in as many levels as in he hiding, ie four Sep Exrac he waermark: In he LL 4 sub band, C conaining he waermark, each coefficien ci, is marked If ci, mod α > α/, hen he recovered binary Figure : Four-Level Wavele Decomposiion 78 P a g e

5 Inernaional Journal of Advanced Technology in Engineering and Science wwwiaescom Volume No, Issue No, November 4 ISSN online: C i, = Ci, Ci, mod α + 75α, if xi, = and Ci, mod α 5α C i, = [Ci, 5α] [Ci, 5 mod α] + 75α, if xi, = and Ci, mod α < 5α 5 C i, = Ci, Ci, mod α + 5α, if xi, = and Ci, mod α 75α C i, = [Ci, + 5α] [Ci, 5α mod α] + 5α, if xi, = and Ci, mod α > 75α bi xi, = ; oherwise xi, = Recompose he binary waermark in he order of he elemens ci, c PSNR is used o evaluae he qualiy beween an aacked image and he original image I is defined as follows: PSNR log db M N fx,y-gx,y M N x y= Where M and N are he heigh and widh of he image, respecively fx, y and gx, y are he values locaed a coordinaes x, y of he original image, and he aacked image, respecively Afer exracing he waermark, he normalized correlaion coefficien NCC is compued using he original waermark and he exraced waermark o udge he exisence of he waermark and o measure he correcness of an exraced waermark I is defined as m n NCC w i, w' i, m n i Where m and n are he heigh and widh of he waermark, respecively wi, and w i, are he waermark bis locaed a coordinaes i, of he original waermark and he exraced waermark IV EXPERIMENTAL RESULTS In his secion some experimens are carried ou o deermine he effeciveness of he proposed algorihm Three hos images are used for he experimens namely Lena, Barbara and Peppers 5 X 5 pixels, 8 bis/pixel 3x3 binary waermark is used Hiding srengh parameer, α was se o 9 if α increases he cover image is he capaciy of he robusness reduces and also percepually he embed image qualiy deoriae The original Lena and i s waermarked images are shown in figures and 3, respecively There was no visual difference beween he original and waermarked images Afer copyrigh proecion, he peak signal o noise raios PSNR beween he original and waermarked images were calculaed and are shown in Table The average PSNR is 4358 db is one of he mos frequenly used formas on he Inerne and in digial cameras The qualiy facor is a number beween and and associaes a numerical value wih a paricular compression level The original and exraced waermarks for some specified qualiy facors are shown in figure 4 for Lena image The Recovery informaion is proved he robusness of he algorihm, even afer he compression wih differen qualiy facor Image PSNRdB Lena 4364 Barbara 436 gold hill 435 Table PSNR values of he hree waermarked images 79 P a g e

6 Inernaional Journal of Advanced Technology in Engineering and Science wwwiaescom Volume No, Issue No, November 4 ISSN online: Figure : Hos image Figure 3: Waermarked image Here he Images are look like same bu he correlaion beween Hos image &Waermarked is no same, means ha informaion is insered in he hos image Figure 4: a original waermark b exraced waermark wihou aack c o e exraced waermarks from compressed Lena waermarked images wih qualiy facors 7, 5, and respecively In his paper, he proposed mehod is compared o Li e al s 6, and Lien and Lin s 6 mehods using he Lena image; he resuls are shown in Table Their copyrigh proecion approaches are blind and heir mehods are based on wavele decomposiion Lena Barbara Goldhill Aacks Li e Lien&lin Proposed mehodncc PSNRdb 46 al NCC Median Filers 3x Median Filers 5x QF= QF= QF= QF= QF=9 78 Sharpening Gaussian Filer3x Gaussian Filer5x5 Average filer3x3 Average filer5x P a g e

7 Inernaional Journal of Advanced Technology in Engineering and Science wwwiaescom Volume No, Issue No, November 4 ISSN online: Table Comparing he proposed mehod wih Li e al s 6 and Lien and Lin s 6 mehod V CONCLUSIONS In his paper, a novel oblivious copyrigh proecion mehod based on he 4-level DWTHaar wavele based informaion hiding algorihm is proposed LL 4 sub band of he hos image is used for hiding he binary waermark Percepual qualiy of he waermarked image is good and he waermark can effecively resis compression wih a qualiy facor greaer han 5 aack The proposed mehod is more robus in resising common aacks such as median filer 3x3 and Gaussian filer and for Compression In addiion o copyrigh proecion, he proposed scheme can also be applied o daa hiding and image auhenicaion his is very useful in mobile communicaion and saellie images o proec he daa because while ransferring he daa from one device o anoher he specificaions may no be same in erms of memory,in ha aspec he daa is compressed obviously even hough he informaion is secured and robus REFERENCES [] J Zhao, Look, i s no here, Bye, January 997,IEEE Trans Circuis Sys Video Technol, vol 3, pp , Augus 3 [] I J Cox, J Killian, T Leighon, and T Shamoon, Secure spread specrum copyrigh proecion for mulimedia, Tech Rep 95, NEC Research Insiue, 995 [3] J Ohnishi and K Masui, embedding a seal ino a picure under orhogonal wavele ransform, in Proc In Conference on Mulimedia Compuing and Sysems, pp 54 5, June 996 [4] R G van Schyndel, A Z Tirkel, and C F Osborne, A digial waermark, in Proc In Conference in Image Processing, vol, pp 86 9, 994 [5] M Kuer, S K Bhaacharee, TEbrahimi, Towards second generaion copyrigh proecion schemes, in Proceedings of he 6h inernaional conference on image processing, Japan, 999 p 3 3 [6] A Nikolaidis, I Pias, Region-based image copyrigh proecion, IEEE Transacion on Image Processing ; :76 [7] J O Ruanaidh, T Pun, Roaion, scale and ranslaion invarian digial image copyrigh proecion, in Proceedings of IEEE ICIP, Alana, GA, 997 p [8] J O Ruanaidh, T Pun, Roaion, scale and ranslaion invarian spread specrum digial image copyrigh proecion, Signal Processing 998; 66:33 7 [9] I J Cox, J Killian, T Leighon, and T Shamoon, Secure spread specrum copyrigh proecion for mulimedia, IEEE Transacion on Image Processing 997; 6: P a g e

8 Inernaional Journal of Advanced Technology in Engineering and Science wwwiaescom Volume No, Issue No, November 4 ISSN online: [] C I Podilchuk, W Zeng, Image adapive copyrigh proecion using visual models, IEEE Journal on Seleced Areas in Communicaions 998; 6: [] A H Paque, R K Ward, I Pias, Wavele packe-based digial copyrigh proecion for image verificaion and auhenicaion, Signal Processing 3; 83: 7 3 [] MVeerli, and JKova_evi_, 995, Waveles and Sub band Coding, Prenice Hall, USA [3] M Hsieh, D Tseng, and Y Huang, "Hiding Digial Waermarks Using Muliresoluion Wavele Transform," IEEE Trans on Indusrial Elecronics, 485: [4] A Reddy and B Chaeri, 5, "A New Wavele Based Logo-copyrigh proecion Scheme," Paern Recogniion Leers, 67: 9-7 [5] P Tay and J Havlicek, "Image Copyrigh proecion Using Waveles," in Proc of he IEEE Midwes Symposium on Circuis and Sysems, pp: 58-6, Oklahoma, USA [6] X Kang, J Huang, Y Q Shi, and Y Lin, A DWT-DFT Composie Copyrigh proecion Scheme Robus o Boh Affine Transform and Compression, IEEE Trans Circuis Sys Video Technol, vol 3, pp , Augus 3 [7]Adii Agrwal,Ruchika Bhadana and Saish Kumar Charan,"A Robus Video Copyrigh proecion scheme using Dw and DCT",IJCSIT,Vol4, [8]DHari Hara Sanosh, Naveen kumar Sarva,Lakshmi Suniha"Robus Video Copyrigh proecion Algorihm using Discree "IJETAEVolume3,Issue5,May3 8 P a g e