Research On Image Digital Watermarking For Copyright Protection And Content Authentication

Digital watermarking as one of effective techniques can be utilized in imagecopyright protection and content authentication. It is also an important supplement fortraditional cryptography and digital signature. In this thesis, three robust watermarkingschemes to improve the robustness are presented firsty. In semi-fragile watermarking,three methods to improve the accuracy of content tamper detection and the quality ofrecovery regions are then discussed.Three robust and three semi-fragile watermarking techniques are presented byusing three types of new transforms and their corresponding coefficient relationships.The main contributions of research are concluded as follows:1. Two robust watermarking algorithms using the Coutourlet and non-redundantCoutourlet transformAn adaptive robust watermarking algorithm in Contourlet domain is first proposed.The significant-tree feature is extracted by using the coefficient relationships and thetexture mask is exploited for improving the tradeoff between the imperceptibility andthe robustness. Comparisons with the classical method in wavelet domain prove theeffectiveness of our proposed method. Another robust watermarking scheme usingWavelet-Based Coutourlet transform (WBCT) is presented in the second algorithm.WBCT is a non-reduandant Contourlet transform and has not applied into watermarkingfield. Through experiments, we find the most of the energy relations between parent andchildren coefficients maintain invariant before and after JPEG compression. Therefore,performance improvement is obtained by means of embedding a watermark exploitingthe modulation of these energy relations. Test results show that our WBCT scheme ishighly robust to different kinds of attacks including non-geometrical and geometricalattacks. These include JPEG 2000 compression (as low as QF=10), 400 pixels circularshifting, and contrast stretching (as low as 20%).2. A semi-fragile watermarking using WBCT for content authentication andlocalizationAfter analyzing the limitation of some of existing methods, such as the security problem, the accuracy of tamper detection and the false alarm rate not mentioned, asemi-fragile watermarking algorithm using WBCT is presented. Furthermore, theauthentication for both content copy and paste attack and other image processingoperations are not resolved in most of the algorithms. Our authenticator is based on thesemi-fragile property of the energy relationship between parent and children coefficients.The relationship not only has an excellent robustness to non-malicious manipulation(JPEG/JPEG2000 compression, additive Gaussian noise, Gaussian filtering and contraststretching) but also provides an effective fragility to malicious manipulations such ascontent copy and paste attack. The watermarking embedding process is accomplishedby the modulation of the parent-children energy relationship. Morphological and fusionoperators are also applied for improving localization accuracy. The degree of theauthenticity is measured by the false alarm and missed detection rates. Simulationresults demonstrated that our proposed algorithm can localize the tampered region withlow false alarm and missed detection rates below 6.50% and 1.50%, respectively, whenboth malicious and non-malicious manipulations are applied together.3. Two semi-fragile methods for tamper detection and recoveryFor improving the quality of recovery image, two schemes are presented. Asemi-fragile, block-wise and content-based watermarking method is presented in thefirst algorithm. For the second method, a pseudo-random binary value watermark isused to resolve the limitation of the false alarm of the tamper detection caused by thewatermark generated by the image block. To improving the number of the recoveredregions, an improved version of 1-D transformation is obtained. In addition, compressedrecovery information is given by 2×2 and 5-bit quantization so that the PSNR (peaksignal-to-noise ratio) values are increased. Experimental results show our method canaccurately detect and localize the tampered region as well as approximately recover itwith the PSNR above 20db.4. Construction of two-dimensional orthonormal filter based on cosine functionfor improving the robustness of digital watermarkingTo resolve the limitation of tensor wavelet in image processing, a cosine functionbased construction method of bivariant wavelet filter is presented. For the proposedmethod, one can easily design different wavelet filters by choosing different references.Experimental results and comparisons with two other filters under the same watermarking algorithm prove the effectiveness of the proposed filter in digitalwatermarking field.