EMPIRICAL MODE DECOMPOSITION IN AUDIO WATERMARKING BY USING WAVELET METHOD

EMPIRICAL MODE DECOMPOSITION IN AUDIO WATERMARKING BY USING WAVELET METHOD G.PAVAN KUMAR (PG Scholor) S.RAHUL (M.Tech) Departmet of ECE, S.V.College of Egieerig ad Techology, JNTUH Assistat Professor, Departmet of ECE, S.V.College of Egieerig ad Techology,JNTUH ABSTRACT: I our proposed framework we preset a ovel watermarkig procedure to embed for copyright protectio ad autheticatio ito digital audio by directly chagig the audio samples the after modifyig the audio sigals. The modified audio sigals are divided ito o. of samples each sample is decomposed adaptively by the method of ovel Empirical Mode Decompositio (EMD).those decomposed samples after decompositio called as a Itrisic Mode Fuctios (IMFs),I this Itrisic Mode Fuctio the low frequecy mode table uder differet attacks is preseted ad the after audio perceptual quality of the origial audio sigal is preserved. Fially i our proposed algorithm we show the robustess efficiecy of the hidde watermark for additive oise re-quatizatio, MP3 compressio,, filterig, croppig ad re-samplig. A compariso aalysis shows that our proposed framework has high ed performace tha the other watermarkig schemes proposed recetly i the literature. INTRODUCTION I digital media the embeddig of watermarkig i audio is for copyrights protectio ad autheticatio.digital media by embeddig a watermark i the origial audio sigal. Mai requiremets of digital audio watermarkig are imperceptibility, robustess ad data capacity. Digital watermarkig has bee proposed as a meas to idetify the ower or distributor of digital data. Watermarkig is the process of ecodig hidde copyright iformatio i digital data by makig small medicatios to the data samples. Ulike ecryptio, watermarkig does ot restrict access to the data. Oce ecrypted data is decrypted, the media is o loger protected. A watermark is desiged to permaetly reside i the host data. Whe the owership of a digital work is i questio, the iformatio ca be extracted to completely characterize the ower. A effective audio watermarkig scheme must satisfy the followig basic requiremets: A. Imperceptibility: The quality of the audio should be retaied after addig the watermark. Imperceptibility ca be evaluated usig both objective ad subjective measures. B. Security: I Watermarkig audio sigals should ot reveal ay clues about the watermarks i them. Also, the security of the watermarkig procedure must deped o secret keys, but ot o the secrecy of the watermarkig algorithm. C. Robustess: After watermarkig extractio Ability to extract a watermark from a watermarked audio sigal after

various sigals processig attacks. D. Payload: The amout of data that ca be embedded ito the origial audio sigal without losig imperceptibility. For audio sigals, data payload refers to the umber of watermark data bits that may be reliably embedded withi a origial sigal per uit of time, usually the extracted iformatio ca be calculated by the BER (Bit Error Rate) Previously differet methods have bee proposed for audio water markig but some problems are arises like robustess, Imperceptibility ad data capacity.now we are proposed a ew algorithm i audio watermarkig for the copyright protectio. That is Empirical Mode Decompositio (EMD). EMD - based time-frequecy aalysis, called Hilbert-Huag Trasform (HHT), this is oly oe of may applicatios made possible by EMD. The fial result ad ideas i time domai applicatios usig EMD apply to two-dimesioal sigals, such as images, as well as audio. EMD decomposes the spatial frequecy compoets ito a set of IMFs (Itrisic Mode Fuctios) where the highest spatial frequecy compoet of each spatial positio is i the first IMF ad the secod highest spatial frequecy compoet of each spatial positio is i the secod IMF, etc. A IMF is defied as a fuctio i which the umber of extrema poits ad the umber of zero crossigs are the same or differ by oe [2]. I the twodimesioal case this demad is relaxed. The upper ad lower evelope of the IMF are symmetric with respect to the local mea, which is used to defie the IMF istead of the umber of extrema poits ad zero crossigs. I two dimesios there are may possibilities to defie extrema, each oe yieldig a differet decompositio. I this work we simply extract the extrema poits by comparig the cadidate data poit with its earest 8-coected eighbors. Approaches, Empirical Mode Decompositio are totally data-drive method that recursively breaks dow ay sigal ito a reduced umber of zero-mea with symmetric evelopes Itrisic Mode Fuctios (IMFs).The startig of decompositio from fier scales to coarser oes. Ay sigal is expaded by EMD as follows: (1) Where z is the umber of IMFs ad (t) deotes the fial residual. The IMFs are orthogoal to each other, ad total IMFs are ear to zero meas. The No. of Extrema is decreased whe ever mode is goig from oe to ext, ad the whole decompositio is guarateed to be completed with a fiite umber of modes. The IMFs are fully described by their local extrema ad thus ca be recovered usig these extrema [7], [8]. Low frequecy compoets which higher order IMFs are sigal domiated [9] ad the their alteratio modes ca lead to degradatio of the sigal. As result, these modes ca be take as to be good locatios for watermark placemet for better robustess. Some results have bee visually i recetly [10], [11] showig the iterest of EMD for audio watermarkig. The EMD algorithm is combied with Pulse Code Modulatio (PCM) ad the watermark is iserted i the sub-bads of a audio which is i trasform domai. Syc-code water mark bits syc- Figure 1: data stretcher Thus the method supposes that mea value of Pulse Code Modulatio audio sigal may o loger be zero. As well as stated by the authors, the method is ot robust to attacks such as filterig,croppig, ad o compariso to watermarkig schemes reported recetly. Our proposed watermarkig is oly based o EMD method ad without domai trasforms, we choosig method a watermarkig techique i the category of Quatizatio Idex Modulatio (QIM) for the reaso to its good robustess ad blid ature.

The Parameters of QIM are selected to guaratee that the embedded watermark is i the last IMF is iaudible. Fially the watermark is associated with a sychroizatio code to facilitate its locatio. Wavelet trasform Wavelet trasform [3] offers effective timefrequecy represetatio of sigals. All basis fuctios are formed by shiftig ad scalig of "mother" wavelet ( t) L 2 R : fuctio m m t 2 2 t 2 m, Z m, Sigal f ( t) L 2 R ca be the represeted as m m, m, f t d t where m, d d are spectral wavelet coefficiets t, f, m m, t For discrete sigals f ( k) L 2 Z hold similar results ad correspodig trasform is called Discrete Wavelet Trasform (DWT). I this article we use orthogoal Haar wavelet trasform, where: 1 for 0 t 0.5 Haar t 1 for 0.5 t 1 0 otherwise A limit of wavelet approach is that the basis fuctios are fixed, ad thus they do ot ecessarily match all real sigals. PROPOSED WATERMARKING ALGORITHM The basic idea of the proposed watermarkig system is to hide the data ito the origial audio sigal a watermark (secret data) with a Sychroized Code (SC) i the time domai format.the iput sigal which is origial audio sigal is first segmeted ito samples after that algorithm EMD is coducted o every samples to extract the associated IMFs (Fig. 1).The all the samples are coverted ito biary data sequece cosisted of SCs ad iformative watermark bits (Fig. 2) is embedded i the extreme of a set of cosecutive last-imfs. All bit (1 or 0) is iserted per extreme. The umber of IMFs ad their umber of extrema deped o the amout of data of each sample; the umber of bits is to be embedded varies from last-imf of oe frame to the followig. Watermark ad Sychroized Code (SC) are ot all embedded i extrema of last IMF of oly oe samples. I geeral the umber of extrema per last-imf (oe sample) is very small compared to legth of the biary sequece to be embedded i audio sigal. This also depeds o the legth of the sample. If we desig by the ad the umbers of bits of SC ad watermark respectively, the legth of biary sequece to be embedded is equal to 2 + The, these bits are spread out o several last-imfs i extrema of the cosecutive samples. This sequece of bits is embedded times i farther. Fially, iverse trasformatio fuctio is applied to the modified extrema to recover the watermarked audio sigal by superpositio of the IMFs of each sample followed by the cocateatio of the sample (Fig. 3). For data extractio presses, the watermarked audio sigal is split ito the o. of samples ad EMD applied to each sample (Fig.1).after that covert ito the Biary data sequeces are extracted from each last-imf by searchig for SCs (Fig. 5).because we are embeddig the data ito last-imf.fig. 6 shows that the last IMF before ad after watermarkig. This figure shows that there is small differece i terms of amplitudes betwee the two modes. The EMD beig full data adaptive, thus it is very ecessary to guaratee that the umber of IMFs will be same as before embeddig the data ad after embeddig the watermark data (Fig. 1),(Fig. 4) respectively However, if the umbers of IMFs are totally differet form origial sigal, there is o guaratee that the

last IMF always cotais the watermark(secret data) iformatio to be extracted. To miimizatio of the problem, the siftig of the watermarked sigal is forced to extract the same umber of IMFs as before watermarkig. The proposed EMD watermarkig scheme is blid, that is, the origial sigal(host sigal) is ot required for watermark extractio. These are the basically 3 steps those are quadrature mirror filters. [1] However i the WPD, both the detail (cd j (i the 1-D case), ch j, cv j, cd j (i the 2-D case)) ad approximatio coefficiets are decomposed to create the full biary tree. 1. Sychroizatio of code 2. Watermark embeddig 3. Watermark extractio Sychroizatio of code: For the hidde the secret data i the origial audio sigal sychroizatio code is used. This code is uaffected (o-degradatio) by shiftig attacks ad croppig [4]. Let P be the origial SC ad Q be a ukow sequece of the same legth. If oly the umber of differet bits betwee P ad Q, whe compared bit by bit, is less or equal tha to a predefied threshold.whe Sequece Q is cosidered as a SC.] WAVELET PACKETS: Origially kow as Optimal Subbad Tree Structurig (SB-TS) also called Wavelet Packet Decompositio (WPD) (sometimes kow as just Wavelet Packets or Subbad Tree) is a wavelet trasform where the discrete-time (sampled) sigal is passed through more filters tha the discrete wavelet trasform(dwt). I the DWT, each level is calculated by passig oly the previous wavelet approximatio coefficiets (ca j ) through discrete-time low ad high pass Fig: Wavelet Packet decompositio over 3 levels. For levels of decompositio the WPD produces 2 differet sets of coefficiets (or odes) as opposed to (3 + 1) sets for the DWT. However, due to the dow samplig process the overall umber of coefficiets is still the same ad there is o redudacy. From the poit of view of compressio, the stadard wavelet trasform may ot produce the best result, sice it is limited to wavelet bases that icrease by a power of two towards the low frequecies. It could be that aother combiatio of bases produce a more desirable represetatio for a particular sigal. The best basis algorithm by Coifma ad Wicker hauser fids a set of bases that provide the most desirable represetatio of the data relative to a particular cost fuctio (e.g.etropy). There were relevat studies i sigal processig

ad commuicatios fields to address the selectio of subbad trees (orthogoal basis) of various kids, e.g. regular, dyadic, ad irregular, with respect to performace metrics of iterest icludig eergy compactio (etropy), subbad correlatios ad others.discrete wavelet trasform theory (cotiuous i the variable(s)) offers a approximatio to trasform discrete (sampled) sigals. I cotrast, the discrete subbad trasform theory provides a perfect represetatio of discrete sigals. AUDIO SIGNALS EMBEBEDED: I the watermark embeddig process Sychroizatio of code are efficietly combied with watermark bits from a obtaied biary sequece Before embeddig, the after it is deoted by {0 1} bit of watermark (Fig. 2). Basics of our watermark embeddig are show i Fig. 3 ad detailed as follows: Steps to be embedded: BLOCK DIAGRAM 1. I step Divide the origial host sigals to the o. of samples 2. Each sample is decomposed ito IMFs (itrisic mode fuctio) 3. Embed T times the biary sequece { } ito extrema of the last IMF ( = ) by QIM Where ad are the extrema of of the origial host audio sigal ad the watermarked sigal respectively. If Sg fuctio is equal to [+] the it is a maxima, ad equal to [-] the it is a miimal. Deotes the floor fuctio, ad S deotes the embeddig stregth chose to maitai the iaudibility costrait 4. Recostruct the samples by usig iverse EMD modified ad cocateate the watermarked frames to retrieve the watermarked sigal. STEPS TO WATER MARK EXTRACTION Host sigal is splitted ito samples ad EMD is performed o each oe as i embeddig for the watermark extractio.extract biary data usig rule give by (3).the fid out the SCs i the extracted data. This procedure is cotiuously repeated by shiftig the selected segmet oe sample at time util a SC is foud. With the positio of SC determied, after that we ca extract the hidde data i.e. iformatio bits, which give as give below Fig.2 watermark embedded presses

Fig. 3. Watermark extractio 1. Divided the watermarked sigal ito o. of samples. 2. Decompose each ad every sample ito IMFs. 3. Extract the extrema of 4. Extract from usig the followig rule [3] SIMULATION RESULTS = 5. Set the start idex of the extracted data, y, to I=1 ad select samples L=N1 (slidig widow size). 6. Evaluate the similarity betwee the extracted segmet V= y(i:l) ad U bit by bit. If the similarity value is, the is take as the SC ad go to Step 8. Otherwise proceed to the ext step. 7. Icrease by 1 ad slide the widow to the ext samples ad repeat Step 6. 8. Evaluate the similarity betwee the secod extracted segmets ad bit by bit. 9. of the ew value is equal to sequece legth of bits, go to Step 10 else repeat Step 7. 10. Extract the P watermarks ad make compariso bit by bit betwee these marks, for correctio, ad fially extract the desired watermark. Watermarkig embeddig ad extractio processes are summarized. Figure: Origial Sigal Figure : Watermarked Sigal

Figure: Watermarked image ad Extracted image Differet attacks No attack Gaussia attack Filterig attack Croppig Attack Figure: IMF of wave files BER(%) 0 0 18.5547 0.1953 Extracted watermark CONCLUSION I our proposed framework Audio sigals which are used for watermarkig techique are divided ito umber of samples each sample is further decomposed adaptively by the method of ew Empirical Mode Decompositio (EMD).those decomposed samples are called as a Itrisic Mode Fuctios (IMFs), I this Itrisic Mode Fuctio the low frequecy table uder differet attacks is preseted ad the after audio perceptual quality of the origial audio sigal is preserved. Fially i our proposed algorithm we show the robustess efficiecy of the hidde watermark for additive oise re-quatizatio, MP3 compressio,, filterig, croppig ad re-samplig.. A compariso aalysis shows that our proposed framework has high ed performace tha the other watermarkig schemes proposed recetly i the literature. REFERENCES [1] V. Bhat, K. I. Segupta, ad A. Das, A adaptive audio watermarkig based o the sigular value decompositio i the wavelet domai, Digital Sigal Process., vol. 2010, o. 20, pp. 1547 1558, 2010. [2] K. Khaldi, A. O. Boudraa, M. Turki, T. Choavel, ad I. Samaali, Audio ecodig based o the EMD, i Proc. EUSIPCO, 2009, pp. 924 928. [3] K. Khaldi ad A. O. Boudraa, O sigals compress EMD, Electro. Lett., vol. 48, o. 21, pp. 1329 1331, 2012. [4] L. Wag, S. Emmauel, ad M. S. Kakahalli, EMD ad psychoacoustic model based watermarkig for audio, i Proc. IEEE ICME,2010, pp. 1427 1432. [5] A. N. K. Zama, K. M. I. Khalilullah, Md. W. Islam, ad I. Molla, A robust digital audio watermarkig algorithm empirical mode decompositio, i Proc. IEEE CCECE, 2010 4