Presenting a Novel Audio Watermarking Based on Discrete Wavelet Transform

Size: px

Start display at page:

Download "Presenting a Novel Audio Watermarking Based on Discrete Wavelet Transform"

Janel Hubbard
5 years ago
Views:

1 Iteratioal Joural of Computer ad Electrical Egieerig, Vol., No., August 2 Presetig a Novel Audio Watermarkig Ba o Discrete Wavelet Trasform Mahdi Mosleh, Saeed Setayeshi, ad Mohammad Mosleh Abstract Digital watermarkig techiques are attractig attetio as a proper solutio to protect copyright for multimedia data. I this paper, we propose a ew audio watermarkig, i wavelet coefficiets i the way that huma auditory system is ot sesitive to it. The propo ca embed the watermark data ito levels - approximatio coefficiets of wavelet trasform usig echo ad discrete Fourier trasform. The extractio algorithm of the propo extracts two watermark sequeces from watermarked data. The fial watermark sequece is obtaied by extracted sequece usig a hidde key. The experimetal results show that the propo is very robust to oise. I order to evaluate, the propo is compared with the s ba o, Echo ad Idex Terms Audio Watermarkig, Discrete Fourier Trasform (DFT) Discrete Wavelet, Trasform (DWT) I. INTRUDUCTION With the recet developmet of digital iformatio techology, all persoal computer owers ejoy a multimedia processor, a vast bad havig access to the etire world ad a iterchageable memory for digital iformatio. Therefore, digital iformatio is accessible to everyoe, ad they are distributed easily. With broadcastig iformatio trasfer by iteret etwork, the firms ca preset their digital products by iteret etwork. Although this techology developmet causes may works to do easy, it has cau some problem as ext techologies. Existig differet devices such as pritig- machies ad copyig media make the copyig of these data easy[]. Now there are two stadard s to protect copyright of digital documets such as codig ad digital sigature. The major problem of codig is that it does ot protect the documet placig i the stage after codig ad is useable. I other words, the user, after applyig certai key ad decodig the documet, ca chage documet i ay way or make uauthorized copies of it i umbers. So, documet ower is ot able to claims his right. If it is made a little chage itetioally or ot i digital documet eve oe bit of it which uses digital sigature to protect copyright, it Mauscript received February, 2; revi July, 2. Mahdi Mosleh is with Departmet of Computer Egieerig, Islamic Azad Uiversity, Dezful Brach, Dezful, Ira ( mahdi_mosleh@yahoo. com). Saeed Setayeshi is with Faculty of Nuclear Egieerig ad Physics, Amirkabir Uiversity, Tehra, Ira ( setayesh@aut.ac.ir). Mohammad Mosleh is with Departmet of Computer Egieerig, Islamic Azad Uiversity, Dezful Brach, Dezful, Ira ( mosleh@ iaud.ac.ir). will be impossible for receiver to retrieve digital sigature. As a result, the documet will lose its credit easily[2]. Studies show that the oly solutio to this problem is to add security iformatio to documet i such a way that it does t separate from documet durig its life time ad also remais udetectable to documet user as much as possible. Oe of the proper solutios is digital watermarkig. Watermarkig is a techique that embeds imperceptible ad statistically udetectable iformatio ito digital data (e.g. video, images ad audio sigals). This embedded iformatio cotais certai iformatio (sigature, logo, ID umber, etc) which uiquely related to the ower or distributer. Digital watermarkig was iveted by oe of the Muzac firm egieers, Emil Hem brook, i. I this ivetio a ID code was added to digital music file such a way that it was udetectable ad imperceptible to proof the owership[]. After that, digital watermarkig was applied i may ways, but this techique did ot attract scietists' attetio as a importat research issue util. At the begiig of, it was attractig attetio as a iterestig research subject which has retaied its importace util ow[]. Durig last decade, differet applicatios of digital watermarkig have bee propo is various fields such as text[], audio[6-], image[, ] ad video[2, ]. These applicatios ca be classified from differet aspects. From of viewpoit the perceptible watermark ito digital medium, two s ca be propo: perceptible watermark ito digital medium [, ]ad imperceptible watermark ito digital medium[2, 6]. Ba o robustess agaist various attacks, watermarkig systems ca be classified ito three categories cosist of robust watermarkig systems[, ], semi-robust watermarkig systems[] ad fragile watermarkig systems[2]. I this paper, we focus o audio watermarkig. Some of the desirable characteristics of a watermarkig algorithm are as follows: iaudibility, robustess agaist sigal processig attacks ad beig udetectable by uauthorized persos. Although various video ad image watermarkig s have bee propo, sice Huma Auditory System (HAS) is more sesitive tha Huma Visual System (HVS), developig a good watermarkig system for embeddig watermark ito audio sigal is more difficult tha image oe. Huma auditory system is able to sese audio distortios with power effect ad low frequecy ad also it is very sesitive to white oise mixig with sigal[2]. Oe of the huma auditory system features is maskig, it meas that oe strog audio removes the auditory effect of aother audio. Huma auditory system is ot sesitive to phase chages ad phase chage distortios are iaudible by ma. These two features (makig ad ot to be sesitive to phase chages)are

2 Iteratioal Joural of Computer ad Electrical Egieerig, Vol., No., August 2 u to develop watermarkig systems[22]. Developig a proper watermarkig algorithm is very difficult. Audio watermarkig s ca be classified ito two mai categories cosist of time domai ba s ad trasform domai ba s. Time domai ba techique is easier to implemet i compariso with trasform domai ba ad eeds less accoutig resources. I other words, time domai ba watermarkig systems are ot robust eough agaist sigal processig attacks i compariso with trasform domai ba systems. [2], Echo[2], s are some kids of techiques ba o time domai. These s are easy to implemet but are ot robust eough agaist attacks. The secod category is s ba o trasform domai which are more robust tha time domai s. Some of them iclude s ba o Fast Fourier Trasform ()[2], s ba o Discrete Cosie Trasform (DCT)[26] ad Discrete Wavelet Trasform (DWT) base s [2] I this paper, we propose a ew audio watermarkig for embeddig watermark sequece ito approximatio coefficiets - levels of DWT, by echo ad s, i order to decrease audibility ad icrease robustess agaist attacks. II. DISCRETE WAVELET TRANSFORM DWT is utilized i a wide rage of DSP applicatios icludig audio/image/video compressio, data commuicatio over the Iteret, patter recogitio ad umerical aalysis[2]. The basic idea of DWT for oe-dimesioal sigals is briefly described. A sigal is split ito two parts, usually the high frequecy (Detail coefficiet) ad the low frequecy part (Approximatio coefficiets). This splittig is called decompositio. The edge compoets of the sigal are largely cofied to the high frequecies part. The sigal is pas through a series of high pass filters to aalyze the high frequecies, ad it is pas through a series of low pass filters to aalyze the low frequecies. Filters of differet cutoff frequecies are u to aalyze the sigal at differet resolutios. Let us suppose that x[] is the origial sigal, spaig a frequecy bad of to π rad/s. The origial sigal x[] is first pas through a half bad high pass filter g[] ad a low pass filter h[]. After the filterig, half of the samples ca be elimiated accordig to the Nyquist s rule, sice the sigal ow has the highest frequecy of π/2 radias istead of π. The sigal ca therefore be subsample by 2, simply by discardig every secod sample. This costitutes oe level of decompositio ad ca mathematically be expres as follows: y [ k] = x[ k]. g[2 k ] high low () (2) y [ k] = x[ k]. h[2 k ] where yhigh[ k ] ad ylow[ k] are the outputs of the high pass ad low pass filters, respectively, after sub samplig by 2. The above procedure ca be repeated for further decompositio. The outputs of the high pass ad low pass filters are called DWT coefficiets ad by these DWT coefficiets the origial image ca be recostructed. The recostructed process is called the Iverse Discrete Wavelet Trasform (IDWT). The above procedure is followed i reverse order for the recostructio. The sigals at every level are up sampled by two, pas through the sythesis filters g [ k] ad h [ k] (high pass ad low pass, respectively), ad the added. The aalysis ad sythesis filters are idetical to each other, except for a time reversal. Therefore, the recostructio formula becomes (for each layer). x[ ] = ( y [ k]. g[ + 2 k] + y [ k]. h[ + 2 k]) high low () To esure the above IDWT ad DWT relatioship the followig orthogoality coditio o the filters H (ω) ad G (ω) must hold: 2 2 H( w) + G( w) = where jw H( w) = h[ ] e ad Gw ( ) = ge [ ] jw III. THE PROPOSED METHOD I the followig of the paper we describe the embeddig ad extractio algorithms of the propo. As usual, audio sigals are aalyzed i short-time form because the audio sigals are o-statioary ad their properties chage durig the time. Therefore, the embeddig ad extractio algorithms are implemeted for each frame. A. Watermark Embeddig Procedure Our propo embeddig scheme has bee show i Fig.. Fig.. Block diagram of watermarked embeddig scheme Watermark sequece embeddig algorithm is explaied i several steps as follows: Step : The calculatio of delay sigal by origial sigal S [ ] DS = SOS d () Step 2: Farmig both origial ad delay sigals. Step: Levels - DWT are obtaied from each frame of origial ad delay sigal. Approximatio ad detail coefficiets of level for both origial ad delay sigal are amed respectively as (CAOS, CDOS) ad (CADS, CDDS). Also, Approximatio ad detail coefficiets of level for origial sigal are preseted as (CAOS, CDOS). Step: Regardig to the watermark data, approximatio coefficiets watermark sigal i level is calculated accordig to Eq. (6). ()

Iteratioal Joural of Computer ad Electrical Egieerig, Vol., No., August 2 ( CAOS ) + coff w = (6) CAWS = ( CAOS ) coff w = where coeff term is a coefficiet betwee [ ].

The approximatio coefficiets level of watermark sigal are calculated by usig the calculated approximatio coefficiets i the former step ad approximatio coefficiet level of delay sigal accordig to Eq.

Watermark extractio procedure Fig.(2) illustrates the propo block diagram for extractig the watermark sequece from watermark sigal. usig the hidde key by two extracted sequeces.

EXPERIMENTAL RESULTS I order to evaluate the propo, two speech ad oe music audio files with wave format were u. These files were 6- bits moo with samplig frequecy 6 KHZ.

The waveforms of a example origial sigal file with its watermarked ad attacked sigals have bee show i Fig.

3 Iteratioal Joural of Computer ad Electrical Egieerig, Vol., No., August 2 ( CAOS ) + coff w = (6) CAWS = ( CAOS ) coff w = where coeff term is a coefficiet betwee [ ]. Also, it is metioable which watermark sequece is shaped by a radom umber geerator usig a hidde key. Step : Iverse wavelet trasform is obtaied from the wavelet coefficiets i former step. The approximatio coefficiets level of watermark sigal are calculated by usig the calculated approximatio coefficiets i the former step ad approximatio coefficiet level of delay sigal accordig to Eq.() : CAWS + coff.cads w = CAWS = () CAWS coff w = Step 6: The iverse wavelet trasform is obtaied from the wavelet coefficiets of watermark sigal. Fially, watermarked sigal frames are calculated. B. Watermark extractio procedure Fig.(2) illustrates the propo block diagram for extractig the watermark sequece from watermark sigal. usig the hidde key by two extracted sequeces. Notice that oly those who have access to this hidde key ca extract the fial watermark sequece IV. EXPERIMENTAL RESULTS I order to evaluate the propo, two speech ad oe music audio files with wave format were u. These files were 6- bits moo with samplig frequecy 6 KHZ. Watermark sequece was formed by a pseudoradom umber geerator usig a hidde key. I all experimets, every audio file was divided ito frames with 2 samples. All implemetatios were doe by Matlab software. The waveforms of a example origial sigal file with its watermarked ad attacked sigals have bee show i Fig.. I order to evaluate the robustess of the propo, seve attacks were applied icludig Additioal Noise (time), Additioal Noise (frequecy), Croppig, Re-quatizatio,, Re-samplig ad Filterig. Bit error rate is calculated by Eq.(): Number of Erroeously decoded bits BER(%) = () Number of embeddig bits for the clip The propo was compared with, Echo ad s. Table (I) shows the experimetal results for two speech audio files (a female ad a male) ad oe music audio file. Also, error rate percetage of, Echo ad s agaist metioed attacks have bee show i Fig.. It should be metioed that S/N ratio, after isertig watermark sequece, i all s is about 2dB. As see from Fig., performace of the propo agaist four attacks Croppig,, Resamplig ad Filterig is too better tha other algorithms. Fig.2. Block diagram of watermarked extractio scheme The watermark sequece extractio process is metioed i some steps as follows: Step: Framig the watermarked sigal ad origial sigal. Step2: Level DWT is obtaied from each frames of watermarked ad origial sigal. Both origial sigal ad watermarked sigal approximatio coefficiets i level are compared with each other accordig to Eq.() ad so the first watermark sequece is extracted. wextracted = if Abs Sum( Abs( CAWS )) Sum( Abs( CAOS )) <. () Step : level DWT is obtaied from each of watermarked ad origial sigal frames. Regardig to the sum of coefficiets differece, accordig to Eq.(), the secod watermark sequece is extracted. w2extracted = Sum real ( (CA WS )) Sum real( (CA OS )) > () otherwise Step : The fial watermark sequece is formed through two extracted watermark sequeces usig user hidde key. I other word, the fial watermark sequece ca be computed Fig.. The wave form of origial sigal watermarked sigal attacked sigal with additioal oise(time) TABLE I. THE OBTAINED RESULTS OF APPLYING DIFFERENT ATTACKS OVER A example male speech A example female speech A example music sigal Additioal. Ech o / 2/2

2/ / / 2/ 2/ 2/ 2/ / Echo (d) Additioal oise(time) Additioal oise (frequecy) Croppig Quatizatio Resamplig Filterig /6 /

Croppig Quatizatio Resamplig Filterig / / / / / / / 6/ / / 6/ /6 / 2/ / /6 /2 /6 2/ 2/ / 2/ 2/ /6 / (g) Fig.

complex i compariso with image ad video watermarkig.

attacks. I this paper, we propose a ew for audio sigal watermarkig i wavelet domai.

4 Iteratioal Joural of Computer ad Electrical Egieerig, Vol., No., August 2 oise(time) Additioal oise (frequecy) Croppig Quatizatio Resamplig Filterig / 2/ 2 / / /6 /2 /6 / 2/ / 22/ 6/ 2/ / / 2/ 2/ 2/ 2/ / Echo (d) Additioal oise(time) Additioal oise (frequecy) Croppig Quatizatio Resamplig Filterig /6 / / / /2 2 / 2/6 / 2/ 22/ 6 6/ 2/ / / 2/ /6 / 2/ /2 / 2/ 2/ /6 (e) (f) Echo Additioal oise(time) Additioal oise (frequecy) Croppig Quatizatio Resamplig Filterig / / / / / / / 6/ / / 6/ /6 / 2/ / /6 /2 /6 2/ 2/ / 2/ 2/ /6 / (g) Fig.. Ifluece over, Echo, ad s Additioal Noise(time) Additioal Noise(frequecy) Croppig (d) Quatizatio (e) (f) Re samplig (g) Filterig V. CONCLUSION Due to dyamics of audio sigal ad also sesitivity of Huma Auditory System (HAS), audio watermarkig is more complex i compariso with image ad video watermarkig. By ow, various techiques have bee reported i this field but most of them are ot eough robust agaist sigal processig attacks. I this paper, we propose a ew for audio sigal watermarkig i wavelet domai. The propo is able to embed a watermark sequece ito two differet levels of wavelet trasform, ba o Echo ad s, ad therefore is c a proper robustess agaist several sigal processig attacks. The experimetal results show that the robustess of the propo is more proper i compariso with, Echo ad s. REFERENCES [] S. J. Lee, ad S. H. Jug, A survey of watermarkig techiques applied to multimedia, i ISIE 2, 2, pp vol..

Iteratioal Joural of Computer ad Electrical Egieerig, Vol., No., August 2 [2] S. Katzebeisser, ad F. A. P.

Memo, Aalysis of LSB ba image stegaography techiques, i 2 Iteratioal Coferece o Image processig, 2, pp. -22 vol.. [] I. J. Cox, ad M. L. Miller, The first years of electroic watermarkig, EURASIP Joural o Applied Sigal Processig, vol.

Bhat K, I. Segupta, ad A. Das, A audio watermarkig scheme usig sigular value decompositio ad dither-modulatio quatizatio, Multimedia Tools ad Applicatios, pp. -. [] H. Kim, Y. Choi, J. Seok et al.

5 Iteratioal Joural of Computer ad Electrical Egieerig, Vol., No., August 2 [2] S. Katzebeisser, ad F. A. P. Petitcolas, Iformatio hidig techiques for stegaography ad digital watermarkig: Artech house, 2. [] R. Chadramouli, ad N. Memo, Aalysis of LSB ba image stegaography techiques, i 2 Iteratioal Coferece o Image processig, 2, pp. -22 vol.. [] I. J. Cox, ad M. L. Miller, The first years of electroic watermarkig, EURASIP Joural o Applied Sigal Processig, vol. 22, o., pp. 26-2, 22. [] Y. W. Kim, K. A. Moo, ad I. S. Oh, A text watermarkig algorithm ba o word classificatio ad iter-word space statistics, Documet Aalysis ad Recogitio, vol. 2, pp., 2. [6] V. Bhat K, I. Segupta, ad A. Das, A audio watermarkig scheme usig sigular value decompositio ad dither-modulatio quatizatio, Multimedia Tools ad Applicatios, pp. -. [] H. Kim, Y. Choi, J. Seok et al., Audio watermarkig techiques, Itelliget Watermarkig Techiques, pp. -2, 2. [] A. Gurijala, ad J. Deller, Robust algorithm for watermark recovery from cropped speech, i IEEE iteratioal coferece o acoustics speech ad sigal processig, 2. [] S. Shi, O. Kim, J. Kim et al., A robust audio watermarkig algorithm usig pitch scalig, i th Iteratioal Coferece o Digital Sigal Processig, 22, pp. -. [] H. Ioue, A. Miyazaki, ad T. Katsura, A image watermarkig ba o the wavelet trasform, i teratioal Coferece o Image Processig,, pp. 26- vol.. [] C. S. Lu, ad H. Y. M. Liao, Multipurpose watermarkig for image autheticatio ad protectio, IEEE Trasactios o Image Processig, vol., o., pp. -2, 2. [2] N. Checcacci, M. Bari, F. Bartolii et al., "Robust video watermarkig for wireless multimedia commuicatios." pp. - vol.. [] F. Hartug, ad B. Girod, Watermarkig of ucompres ad compres video, Sigal Processig, vol. 66, o., pp. 2-,. [] M. D. Swaso, B. Zhu, B. Chau et al., Object-ba trasparet video watermarkig, i IEEE First Workshop o Multimedia Sigal Processig,, pp. 6-. [] M. D. Swaso, B. Zhu, ad A. H. Tewfik, Trasparet robust image watermarkig, i Iteratioal Coferece o Image Processig, 6, pp. 2-2 vol.. [6] T. Furo, ad P. Duhamel, Robustess of asymmetric watermarkig techique, i Iteratioal Coferece o Image Processig, 2, pp. 2-2 vol.. [] P. J. Lee, ad M. J. Che, Robust error cocealmet algorithm for video decoder, IEEE Trasactios o Cosumer Electroics, vol., o., pp. -,. [] M. Ramkumar, ad A. N. Akasu, A robust protocol for provig owership of multimedia cotet, IEEE Trasactios o Multimedia, vol. 6, o., pp. 6-, 2. [] D. He, Q. Su, ad Q. Tia, A semi-fragile object ba video autheticatio system, i Iteratioal Symposium o Circuits ad Systems, 2, pp. III--III- vol.. [2] J. Fridrich, M. Golja, ad A. C. Baldoza, New fragile autheticatio watermark for images, i Iteratioal Coferece o Image Processig, 2, pp. 6- vol.. [2] F. Hartug, ad M. Kutter, Multimedia watermarkig techiques, i Proceedigs of the IEEE,, pp. -. [22] L. Boey, A. H. Tewfik, ad K. N. Hamdy, "Digital watermarks for audio sigals." p.. [2] W. N. Lie, ad L. C. Chag, "Robust ad high-quality time-domai audio watermarkig subject to psychoacoustic maskig." pp. - vol. 2. [2] W. Beder, D. Gruhl, N. Morimoto et al., Techiques for data hidig, IBM systems joural, vol., o. /, pp. -6, 6. [2] M. Fallahpour, ad D. Megias, High capacity audio watermarkig usig amplitude iterpolatio, IEICE Electroics Express, vol. 6, o., pp. -6, 2. [26] X. Y. Wag, ad H. Zhao, A ovel sychroizatio ivariat audio watermarkig scheme ba o DWT ad DCT, Sigal Processig, IEEE Trasactios o, vol., o. 2, pp. -, 26. [2] M. Pooya, ad A. Delforouzi, Adaptive ad robust audio watermarkig i wavelet domai, i Third Iteratioal Coferece o Iteratioal Iformatio Hidig ad Multimedia Sigal Processig, 2, pp [2] S. G. Mallat, A wavelet tour of sigal processig: Academic Pr,. Mahdi Mosleh, received his B.S. i computer egieerig from Islamic Azad Uiversity, Dezful Brach, i 2 ad the M.S. i computer egieerig from Islamic Azad Uiversity, Dezful, i 2 i computer egieerig. His research iterests are Audio ad Image watermarkig. Saeed Setayeshi, B.Sc., M.Sc., M.A., M.A.Sc., Ph.D. (Electrical & Computer Eg., TUNS, Caada, ) is a Associate Professor ad teachig i the Faculty of Nuclear Egieerig ad Physics, Amirkabir Uiversity of Techology (Tehra Polytechics). He is also presetig some courses i Computer Eg. Departmet of Res. & Sc. Brach of IAU as a Ivited Professor. He has fouded AL ad Complex System Researches for first time there ad supervi may Graduate studets i the areas of usig CA ad LA. He has published more tha papers i ISI Jourals ad Cofereces. His research iterests are i the areas of AI & AL, Itelliget Cotrol (Neural Fuzzy Expert-GA-CA-LA), Adaptive Sigal Processig, Aget Ba Modelig, Artificial Society, Social Evolutio, Wealth Distributio, Kowledge Ba Systems, ad Dyamics of Complex Systems. Mohammad Mosleh, received his B.S. i computer egieerig from Islamic Azad Uiversity, Dezful Brach, i 2, the M.S. i computer egieerig from Islamic Azad Uiversity, Tehra, i 26 ad the PhD degree i computer egieerig at the Islamic Azad Uiversity, Tehra, i 2. He is assistat professor i the Departmet of Techical ad Egieerig at the Islamic Azad Uiversity, Dezful Brach. His research iterests are i the areas of Speech Processig, Machie Learig ad Parallel Processig.

Novel Steganography System using Lucas Sequence

Novel Steganography System using Lucas Sequence Novel Stegaography System usig ucas Sequece Fahd Alharbi Faculty of Egieerig Kig Abdulaziz Uiversity Rabigh, KSA Abstract Stegaography is the process of embeddig data ito a media form such as image, voice,