Watermark-based authentication and decentralized recovery techniques for multimedia content preservation

The information age has brought unprecedented benefits to society, but at the same time, it has brought enormous technical challenges. The preservation of online digital content is one such formidable challenge. Preservation refers to safeguarding and providing continual access to document contents in their authentic original form, and document authentication procedures are an essential aid in ensuring preservation. Recently several commercial forums and government consortiums have been established to identify requirements for the preservation of online digital contents and the establishment of related standards. Most of the existing research efforts related to online content preservation have been aimed at preserving textual documents. Such techniques when applied to multimedia contents (video, image, audio) fail to address the issues peculiar to multimedia. Moreover, these efforts assume a managed networked environment where a consortium of trusted publishers or users collaborate together to preserve and authenticate the originality of a document.;This dissertation investigates the design and development of a preservation system that specifically caters to the needs and structure of multimedia data. It exploits the perceptual nature of multimedia content, and proposes an authentication scheme that is based on watermarking the contents using content-based signatures. The major contributions of this dissertation include: (i) an efficient hashing based procedure to generate signatures that capture local and global document information, (ii) multiple Compensated Signature Embedding (CSE) schemes to robustly embed the signatures in the media as a watermark, such that perturbations introduced in the media due to embedding are compensated, and (iii) a highly scalable decentralized recovery procedure to recover the tampered/corrupted media. While technically a highly challenging endeavor, the proposed preservation system alleviates the explicit need for a third party authentication agent. For CSE schemes, both iterative and closed-form solutions have been devised. The iterative schemes employ set theoretic approach of Projections Onto Convex Sets (POCS) and additive compensation, and the closed-form scheme incorporates the Lagrange-multiplier based optimization. The proposed data recovery procedure exploits embedded signature information to identify the corrupted or tampered portions of the media and polls its peers in the network to recover the media, reflecting its applicability to preservation of multimedia contents in decentralized P2P networks.