Digital Watermarking And Shape Correspondence Of3D Models

With rapid development of acquisition facilities and processing techniques, three-dimensional (3D) models nowadays are widely applied to digital entertainment, film and television production,3D games, cultural heritage protection, etc. How to protect copy-rights of3D models effectively and prevent unauthorized duplication, modification, and spread is becoming an important problem in digital geometry processing. The digital wa-termarking techniques for3D models based on information hiding theory are proposed accordingly, which provide an effective means of copyright protection and ownership as-sertion.The thesis focuses on robustness and confirmability of watermarking algorithms for3D mesh models. By optimally determining a center point set based on quadric error mesh simplification, a robust spectral watermarking algorithm based on radial basis functions is proposed. Inspired by digital signature algorithm and manifold harmonics analysis, a prac-tical framework of robust confirmable watermarking algorithm for two-manifold meshes is presented. Based on global point signature and Green’s function representation, a coarse-to-fine shape correspondence algorithm for nearly-isometric meshes is proposed. The de-tails are as follows:To improve robustness and efficiency of3D mesh watermarking algorithm based on radial basis functions, a new robust watermarking algorithm for large scale3D meshes is proposed. Firstly, by adopting an optimized center point set determined via quadric error mesh simplification, the proposed algorithm decreases singularity of pre-basis function-s and improves the mesh reconstruction quality. To resist cropping attack and improve watermarking robustness, the large scale3D mesh is semantically segmented into several patches, and the signature bits are embedded into the low-frequency spectral coefficients absolutely at a proper starting position. In extraction step, the signature bits are extract-ed weightedly with regard to their corresponding disturbed spectral coefficients. Finally, through detailed and comprehensive comparisons with other spectral watermarking algo-rithms, it shows that the proposed algorithm exhibits better computational efficiency and robustness against to various attacks.Based on manifold harmonics analysis, a robust confirmable spectral watermarking al-gorithm for a two-manifold mesh is proposed, where ownership of the watermarked mesh can be confirmed by a trusted third party. Without sacrificing of invisibility and robustness of the3D watermarked mesh, we give and prove formula of the minimal number of mani-fold harmonics basis functions required in the proposed algorithm. It greatly facilitates the spectral analysis and watermark embedding of a large-scale3D mesh containing millions of vertices. Then, a digital signature algorithm is incorporated explicitly into both water-mark embedding and extraction phases, which optimizes non-blind spectral watermarking theory framework and addressed confirmability of3D watermarked meshes. To improve the robustness of the algorithm, the watermark signature bits are embedded in an absolute manner. Extensive and thorough experiments show that the proposed algorithm exhibits better invisibility and is more robust to resist various geometry and connectivity attacks than existing watermarking algorithms. Meanwhile the distributions of extraction rates and the receiver operating characteristic curve show the validity of the algorithm also.Based on global point signature and Green’s function representation, a coarse-to-fine shape correspondence algorithm for nearly-isometric meshes is proposed. First, to address the problem of undetermined signs of corresponding basis vectors, the anchor points are ex-tracted from the source mesh and the nearly-isometric destination mesh respectively based on the Morse theory and a modified hierarchically clustering approach. Their correspon-dences are then established by minimizing their GPS distances, which can be solved via a combinational searching strategy. Then, to solve the inconsistence of high dimensional GPS coordinates caused by different resolutions of meshes, a new Green’s function presen-tation of point is defined based on the anchor point pairs obtained. And an incremental algo-rithm is proposed subsequently for sparse correspondence. Finally, dense correspondences are established through minimizing the Euclidean distance defined by Green’s function representation and anchor points. Experimental results on the benchmark data set demon-strate the proposed algorithm exhibits better computational efficiency and correspondence accuracy than other shape correspondence algorithms. The algorithm provides the founda-tion for designing watermarking algorithm to against nearly-isometric shape deformation attacks, and can be potentially applied to rigid or non-rigid mesh alignment,3D morphing, shape matching, etc.Finally, we conclude the thesis and discuss some directions for future work.