Abstract

Since their introduction, photo sharing and social network websites such as Flickr, RockYou, MySpace, PhotoShelter, and FaceBook have attracted millions of users, many of whom are using these sites to share digital photos, but there is an increasing concern about controlling copyrighted content. Using digital image watermarking technology, online photo-sharing users will not only be able to monetize nonlinear distribution of copyrighted content, but also apply rules to how the content can be used and shared in order to avoid having the digital photos corraled and reused. This thesis introduces a robust invisible image watermarking mechanism using nonnegative matrix factorization and singular value decomposition approaches. The proposed watermarking scheme improves the performance of the data embedding system effectively, and provides resistance to a wide range of attacks. Illustrating experimental results show the much improved performance of the proposed watermarking approach in comparison with existing techniques, and also demonstrate its robustness against a variety of intentional attacks and normal visual processes. Inspired by the successful application of graph theory in 2D image watermarking, and looking to address its limitations in often difficult 3D model identification and authentication, we also present in this thesis a 3D mesh fingerprinting technique using spectral graph theory. The main idea is to partition a 3D model into sub-meshes, then apply eigen-decomposition to the Laplace-Beltrami matrix of each sub-mesh, followed by hashing each transformed sub-mesh. The experimental results using a benchmark of 3D models demonstrate the effectiveness of the proposed fingerprinting technique