MULTIPURPOSE INVISIBLE AND VISIBLE DIGITAL IMAGE WATERMARKING SCHEMES

Abstract

Watermarking is an advanced research topic in multimedia community. It is used for various applications such as broadcast monitoring, piracy, owner identification, copyright protection, copy deterrence, proof of ownership, media authentication, fingerprint/transaction tracking, copy control, legacy enhancement, security of biometric system etc. The requirements of an efficient watermarking system depend on the application scenario. Nowadays, the main research issues in watermarking are development of new watermarking schemes, developing the testing tools for new and developed watermarking schemes, analysis of a watermarking system and broaden the usefulness of watermarking to multimedia applications. In this thesis, we have examined the solutions of five watermarking related problems. The first problem has been discussed in Chapter 2. Here, the digital image watermarking and biometrics have been combined to improve the owner identification/verification technology. This problem is related to broaden the watermarking application. For the solution, we have developed four watermarking schemes. Two watermarking schemes operate on discrete wavelet transform domain and rest two schemes operate on redundant discrete wavelet transform domain. One watermarking scheme in each transform domain incorporates a weighted binary coding. We have observed that the discrete wavelet transform based watermarking scheme coupled with the weighted binary coding provides the best solution for the first problem. i ii Both non-blind and blind watermarking schemes in the real oriented wavelet transform domain are proposed in the second problem (chapter 3). One challenge in this problem arises due to the fact that the product of the left inverse of real oriented wavelet transform followed by the real oriented wavelet transform (ROWT) is not the identity transform. A key contribution towards the solution is that we have obtained a product of left inverse of ROWT followed by the ROWT itself. This product has been used in both the proposed schemes. Further, derived mathematical properties based on the quotient-remainder-theorem have been used in the proposed blind watermarking scheme. The third problem (Chapter 4) is to provide a solution to improve the security/integrity of a compromised biometric system. In the solution of the third problem, we have proposed a spatial domain and a discrete cosine transform based watermarking schemes. The spatial domain based watermarking scheme is blind, reversible and fragile while the discrete cosine transform based watermarking scheme is blind and robust. The fourth problem (Chapter 5) is motivated by the fact that visible watermark is embedded at predefined positions in an image/video frame, which leads occlusion of important portion of multimedia objects by the visible watermark, for instance, TV channel logo used for television broadcasting. As a solution, we have proposed a visible watermarking technique that embeds a binary logo watermark at N non-overlapping positions in an image such that important portions of the image are not occluded. In the last problem of the thesis (Chapter 6), we have analysed watermarking systems of binary watermarks. The analysis is based on an assumption that negative paired images provide same information, which is the basic fundamental of information theory. We have analysed watermarking systems with respect to five comparators. These five comparators are based on the normalized Hamming iii similarity (NHS), the normalized correlation coefficient (NCC/NC), the mean subtracted NCC (MSNCC), symmetric NHS (SNHS, a version derived from NHS), absolute MSNCC (AMSNCC, a version derived from MSNCC). A main observation of this analysis is that SNHS and AMSNCC based comparators treat negative of watermark same as itself. The solution of each problem has been tested against several signal distortions such as Geometric attacks, filtering, format conversion and noise addition.