DIGITAL WATERMARKING FOR IMAGE AUTHENTICATION AND RECOVERY

Abstract

Revolution in generation, storage, and communication of digital information has brought about profound changes in our society. The digital information age has evolved with numerous opportunities and new challenges. The goal of this thesis is to provide a framework based on watermarking techniques that can be used for verifying the integrity of a cover work along with the recovery of the intentionally/unintentionally distorted cover work. In this context, multiple semi- fragile watermarking techniques are proposed for not only protecting the digital content from alteration but also to recover it after alteration. In these techniques, watermarks remain intact with the image under minor enhancement and is broken only in case of major manipulations. In this thesis, a hybrid scheme is developed that can both authenticate and recover the altered image. The first phase of this thesis encompasses an improvement of a self-recovery authentication scheme for digital images. The second phase then considers the reduction in computational complexity. Finally, a novel model is proposed in the third phase that not only perform accurate authentication of images but also recovers the altered image. In this thesis, the concept of multiple watermarking is employed; authentication and recovery watermarks. Both of these independent watermarks strengthen the security aspect of each other and it is user choice to use both of the watermarks or one of them according to the requirement of application. The authentication watermark is correlated to the host image for resisting collage attack and then embedded in the wavelet subbands. Unlike the conventional block-based approaches, it has the ability to determine the regions concisely where the integrity verification fails. The recovery watermark recovers the image with original quality even after manipulation of the watermarked image. Lossless compression (Huffman coding) and BCH (Bose, Ray-Chaudhuri, and Hocquenghem) coding are utilized while generating the recovery watermark. Integer DCT is utilized instead of conventional DCT because the integer DCT contents can be highly compressed by Huffman coding. In addition, integer wavelet transform, which is a fast approach of discrete wavelet transform, also reduces the computational complexity of the proposed algorithm. In contrast to the earlier authentication algorithms, the proposed techniques exploit flexibility in both of the watermarks, where a trade-off can be made by the user according to the requirement of application. Experimental investigations are performed to evaluate the performance of multiple semi-fragile watermarks. It is demonstrated that the performance of the proposed methods is better compared to the conventional block-based approaches in context of tamper detection. In summary, efficient techniques have been developed in this thesis which makes trade-off between three contradicting properties of watermarking; imperceptibility, robustness, and capacity. The proposed watermarking techniques are able to answer these questions, i) Has the image been processed? ii) Has the image been processed incidentally or maliciously? iii) Which part of the image has been processed and how much? Additionally, a self-recovery approach makes it possible to recover the exact version of the host image even after the image has been incidentally/maliciously processed.