| Example-based texture synthesis technology is hot pursuit in recent years and playing an important role in the areas of computer vision, image processing and computer graphics, and it is leading a great prospect. The previous methods rely on a single input image that can capture a limited band of spatial scales and create a similar texture perceptually. However, synthesizing a continent-like appearance at a variety of zoom levels would require a high input resolution that can't be obtained in practice.At present, there is a multi-scale texture synthesis method, which needs to input multiple exemplars, construct data structure such as an exemplar graph and Gaussian stacks with all exemplars and then synthesize the final multi-scale texture through a series of the sampling, jitter, and correction techniques, in the two-dimensional field of texture synthesis. The approach in this article is proposed based on it and extended to the three-dimensional field. So a new technology, three-dimensional multi-scale texture synthesis, is put forward. This solves the trouble in the field of texture synthesis.In this paper, the concept and measurement of self-similarity about two-dimensional textures are introduced. And two methods about the strength of the self-similarity are researched. This provides foundation for the reasonable use of the method presented in the paper. And then the theory and implementation of the two-dimensional multi-scale texture synthesis are discussed in detail. Then the application of the technology is explored in the fractal texture and the gradient texture synthesis. This is a bright spot in this article. Finally, the thesis makes a detailed introduction to the innovation that the two-dimensional multi-scale texture synthesis method is extended to the three-dimensional field. A large number of experimental results are placed in order to prove the feasibility of it. |