基于GPU加速的几何纹理合成方法

提出了一种基于GPU加速的几何纹理合成方法，以解决几何纹理合成过程中高计算量、高存储占用和高耗时等问题.首先，对样本几何纹理数据进行子块划分，并根据子块在样本中的位置关系设计可重用样本顶点数据的数据结构，优化存储以降低内存的占用率；然后，采用GPU多线程并发技术设计并行加速算法，将串行的几何纹理合成过程并行化，从而实现快速生成任意尺寸的新的几何纹理.实验结果表明，该算法不仅占用存储较少，而且在保证合成质量的同时极大地降低了几何纹理的合成耗时.

GPU-based geometry texture synthesis

A geometry texture synthesis method based on GPU technique is proposed to solve the problems of high computation, high memory occupancy, and high time consuming in synthesis process. Firstly, the geometry texture sample can be divided into sub-blocks, and the data structure reusing the vertex data of the geometry texture sample is designed according to the positions of these sub-blocks in the geometry texture sample, in that the storage optimization can reduce the memory occupancy rate. Then, based on multithreaded GPU technique, we design parallel acceleration algorithm, and develop the sequential geometry texture synthesis in parallel, which can generate a new synthesized geometry texture with arbitrary sizes fastly and efficiently. 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 and guarantee the quality of the synthesize geometry texture.