| In virtual reality,texture data is usually apply to simulate the surface details of model and scene,which the realistic of model and the scene is markedly improved.Texture can be divided into 2D texture and geometric texture according to the data structure of the texture.The 2D texture is mainly based on the two-dimensional image data with self similarity,2D texture synthesis is provided with high synthetic speed and memory footprint small;at the same time,due to the 2D textures are not supported on occlusion,shadow and outline these important effect,2D texture unable to meet requirements of virtual reality to form more and more refined.The geometric texture has more abundant and detailed performance on the surface of model and scene,has been widely used in virtual reality technology,greatly improve the realism of models and scenes.Large amount of calculation,high storage occupancy and a large number of manual interaction and so on This is particularly a problem for existing geometric texture synthesis method,So how to synthetic high-quality geometric texture efficiently,has become a hot research topic in computer graphics.First,the thesis describes the research background,research significance and study status at home and abroad of the texture synthesis.And some of the classic algorithm of 2D texture synthesis and geometric texture synthesis are introduced and analyzed in detail and compares the advantages and disadvantages.Then,the Markov model,neighborhood pixel,texture block neighborhood and neighborhood matching principle of texture synthesis basic theory are described in detail,and texture synthesis general steps are described in detail combined with the existing classical methods of 2D texture synthesis and geometric texture synthesis.In order to analysis influence the speed of texture synthesis method which based on neighborhood matching,is a large number of neighborhood search became the bottleneck of the speed of texture synthesis.Secondly,the texture synthesis accelerated technology are introduced from the aspects of software and hardware in detail,according to the bottleneck problem of neighborhood search,this paper deeply studied related software acceleration technology and GPU multi thread parallel acceleration techniques.A geometry texture synthesis method based on GPU technique is proposed to solve the problem of high computation,high memory occupancy,and high time consuming in synthesis process.We design parallel acceleration algorithm based on multithreaded GPU technique,developing the sequential geometry texture synthesis in parallel The experimental results show that our algorithm not only can use less memory,but also can greatly reduce the time consuming for geometric texture synthesis,guaranteeing the quality of the synthesize geometry texture. |
