Texture Synthesis From Samples Over Point Set Surface

For the wide applications in TV and file producing, digital entertainment, industrial design and virtual simulation, etc., the research on texture has been one of the hottest topics in computer graphic, vision and image processing. Texture Synthesis from Samples (TSFS) is a new texture tiling technique following the Texture Mapping and Procedural Texture Synthesis. Many techniques for 3D Texture Synthesis from Samples have been put forward till now, but their application to synthesize textures is impossible in the case of complicatedly structured object. Based on 3D mesh surface, these traditional techniques are to render according to Tri Patches made by point set. With the high-speed development of 3D scanning method, the number of mesh vertices gotten from scanning has reached one billion or more. So in the face of over-million-size mesh models, traditional mesh simplification algorithm has lost its practicability due to the limitation of time and space. Abandoning the traditional way of rendering for Tri Patches, Point-Based Rendering only record points message on which the final image is rebuilt, thus offering a new approach to fast rendering of Point Set with large amounts of points.Up till now, with the development of geometry modeling and rendering over Point Set Surface, Point-based rendering is getting closer to maturity. In this thesis, a new 3D texture synthesis technique based on Point Set Surface is put forward. The working flow of this approach is presented: (1) the direction fields are built over point set surface ; (2) an initial point for texture synthesis is selected; (3) neighborhoods for every point are built ; (4) the texture of the initial point is obtained by selecting texture from the input image randomly, at the same time all neighborhood points of the initial point are found and theirs textures are further initialized; (5) enter the process of texture synthesis and in each circulation select the next point for texture synthesis from point set surface; (6)the mesh neighborhood construction is obtained; (7) the most matching point is found in the input sample image; (8) the texture of the most matching point is transferred to the point which is selected from point set surface.The new 3D texture synthesis technique in the thesis is realized through algorithms. There are several controllable parameters in the algorithms, such as the number of points in the mesh neighborhood, the unit distance of mesh neighborhood construction, the number of points in kdtree. These parameters are dynamically changeable, thus could realize the controllable quality of the algorithms.Using this new approach, we synthesize texture on Point Set Surface using different input image, and effective result is achieved for each input image. The experimental results show that the new technique can produce smooth texture synthesis effects as well as keep the structures in the sample texture on the point-sampled geometry. Thus, the new technique is effective and applied for 3D texture synthesis. In contrast with traditional 3D texture synthesis, the technique put forward in this thesis uses less time and is more flexible and more controllable and of robustness.