Research On Digital Watermark Methods For 3D Mesh Models

With the rapid development of virtual reality, cooperative design and Web 3D, more and more companies or copyright owners present their 3D digital media product through Internet. The characteristic that digital media product can be copied with no distortion, transported through Internet with high speed has benefited them a lot. But meanwhile it also caused the problem of unauthorized using. Illegal use of copyrighted material can be prevented by embedding a licensee's identity into data using digital watermark method. The copyright owner can then identity the source from which illegal copied spread, or proves ownership in disputes.The thesis proposes two algorithms according to two kinds of application of 3D model watermark, one is used for private using local set information, and another is for public using principle direction method. The contribution work of this thesis is shown as following:Firstly, we propose an algorithm to watermark 3D objects using local set information. In this method, we choose vertices in rapid change area as embedding points. For very embedding vertex, we compute the normal vector of local set of those vertices. The watermarked vertices are moved to lie in different direction to hide different watermark information. In detection, the same embedding vertices are selected base on their local set area information. The watermark information is attracted according to which direction the test vertex lies in. The experiment result has demonstrated that this method can be good used in private watermark application.Secondly, we propose another algorithm to watermark 3D objects using principle direction method. In the embedding process, we divide the embedding space into eight in the way of quantization watermark method. The watermarked vertices are moved to lie in those subspaces to represent different watermark information. As the detection progresses, the same vertices area information is got to build the synchronization between embedding and detecting. The statistical invariant property of first order moment makes blind detection can be realized. The experiment has shown that, while the attacks aren't as strong as to damage the visual quality of 3D object, the algorithm acquires high robustness to resist attacks of adding noise and smooth-filtering. It can perfectly satisfy the public watermark application.