A robust digital watermarking using discrete wavelet transform in chrominance channel

The obtained results show significant improvement in terms of quality and fidelity of extracted signal. Also, the improvements in terms of robustness by the proposed method work effectively against various common image-processing-based attacks especially against compression attacks.
Nội dung trích xuất từ tài liệu:
A robust digital watermarking using discrete wavelet transform in chrominance channelJournal of Automation and Control Engineering, Vol. 1, No. 2, June 2013A Robust Digital Watermarking using DiscreteWavelet Transform in Chrominance ChannelKomwit SurachatInformation and Communication Technology Programme, Faculty of Science, Prince of Songkla University, ThailandEmail: komwit.s@psu.ac.thAbstract—A wavelet-based digital watermarking scheme ispresented in this paper. The watermark signal is embeddedinto HLn of LLn-1 transformed sub-band of originalchrominance channel host image using coefficients scalingtechnique. In the extraction process, the average filter isapplied to predict original dwt coefficients in HLn-1 subband. Then, the watermark logo is recovered by subtractingthe coefficients predicted from the embedded coefficients,without original image needed. The obtained results showsignificant improvement in terms of quality and fidelity ofextracted signal. Also, the improvements in terms ofrobustness by the proposed method work effectively againstvarious common image-processing-based attacks especiallyagainst compression attacks. example, M. K. Samee et al. [4] presented reversiblewatermarking scheme for images by using CMDA basedin wavelet domain. In addition, [5] and [6] proposedwatermarking algorithms in frequency domain usingdiscrete wavelet transform. They applied discretefractional Fourier transformation DWT and Region ofInterest (ROI) technique, respectively.In this paper, we present a wavelet-based digitalwatermarking scheme by applying some spatial domaintechniques to improve the performance of extractedwatermark signal. The coefficient scaling method isproposed to adjust area before embedding which will beexplained more in section 2. Then, in section 3, theexperimental results are shown and discussed. Theconclusion is finally drawn in section 5.Index Terms—discrete wavelet transform,digitalwatermarking, chrominance channel, image processingbased attackI.II.This proposed watermarking scheme consists of twoprocesses those are the embedding process and theextraction process. In addition, we also present sometechniques to transform the color channel and the sizes ofwatermark image and host image before performing theoperations. The detail of the embedding process and subprocesses are given next.INTRODUCTIONThe copyright of digital media protection becomes animportant topic nowadays, due to one main characteristicof digital multimedia which is the ease of copying andredistributing without losing quality. To resolve thecopyright protection problem for multimedia data, manywatermarking schemes are proposed and discussed.We can classify watermarking methods into frequencyand spatial domain based watermarking. In the spatialdomain, watermark embedding can be accomplishedeasily by modifying the image pixels directly. Forinstance, [1], [2] and [3] also proposed the watermarkingschemes in spatial domain by using spatial correlation ofcolors, RST-resistant method and independent componentanalysis, respectively. However, there are manyarguments about robustness against compression attackse.g. JPEG2000 and JPEG compression standard. Incontrast, in the frequency domain based approach, it isobvious that the robustness against compression attack ismuch better than spatial domain based.Moreover, frequency domain watermarking stronglyhelps increasing the imperceptibility, security, androbustness. Therefore, presently, most of imagewatermarking methods are in the transform domain. ForA. Pre-processing Operation of Host ImageThe host image is first pre-processed beforeembedding with the following steps;Step 1: The original RGB host image I with n by npixels is converted to YCbCr color space, the chrominancechannel, Cr is selected to use in this process.Cr  Cr( i , j ) | 1  i  n,1  j  n,16  Cr(i , j )  240Then, the host image Cr is decomposed into n levelsusing discrete wavelet transform. The HLn of LLn-1transformed sub-band is selected for watermarkingembedding.an , qn , bn , rn   dwt LLn1 (1)where a n , qn , bn , rn is wavelet coefficient value ofLLn ,HLn LHn and HHn sub-band, respectively.Step 2: The coefficient values qn(i,j) is scaled between0-255 by using following equations;Manuscript received October10, 2012; revised December 22, 2012.This research work is supported by the Information andCommunication Technology Programme, Faculty of Science, Prince ofSongkla University.©2013 Engineering and Technology Publishingdoi: 10.12720/joace.1.2.99-102PROPOSED WATERMARKING SCHEMEqn(i, j )   minqn   qn(i , j )99(2)Journal of Automation and Control Engineering, Vol. 1, No. 2, June 2013 255 qn(i , j )  qn(i , j )  max(qn ) (3)where q’n(i,j) is the scaled coefficient value at position (i,j),min(qn) is the minimum value of HLn sub-band waveletcoefficient and max(qn) is the maximum value of HLnsub-band wavelet coefficient.B. Pre-processing Operation of Watermark SignalThe watermark signal is first pre-processed with thefollowing steps;Step 1: watermark pixels are converted from {0,1} to{1,-1} by switching the value of the zero bits to the onebits.D. The Extraction ProcessStep 1: RGB color model is transformed to YCbCrcolor model. Next, the transformed Cr is decomposed inton levels using discrete wavelet transform. Then, thescaling equation is applied to adjust the coefficients intothe right proportion the same as the embedding process.Step 2: Each original transformed image pixel in thechosen sub-band is predicted from its surrounding scaledcoefficient values. The predicted original image pixelq’’’(i,j) is calculated byq ( i , j ) 19im  1jn  1q ( i  m , j  n )(7)Step 3: The embedded watermark bit w’ (i,j) at a givencoordinate (i,j) can th ...