使用特征点定位的纹理优化

Texture optimization using feature point locating

提出一种纹理优化算法,以加速纹理合成和确保输出图像结构的正确.在预处理阶段,分析纹理样图中特征点的全局分布,并计算样图结构坐标.在优化阶段,像素邻域匹配之前通过结构坐标匹配排除会导致图像结构错误的匹配位置,避免不必要的匹配计算并确保图像结构的正确;使用低能量的初始值及结构坐标动态调整加速迭代过程的收敛.实验证明,该算法提升了图像质量并加速了合成过程,加速比随纹理随机度的不同而不同,对大多数结构性纹理,至少达到两个数量级.

In order to speed up texture synthesis and assure the correctness of output image structure, a texture optimization algorithm was presented. In the preprocessing stage, the global distribution of feature points in the sample texture was analyzed and structural coordinates of the sample texture were computed. In the optimization stage, structural coordinate matching was performed before pixel neighborhood matching to get rid of matching positions that caused image structure error. This avoided unnecessary matching computation and assured correctness of image structure. Low energy initial value and dynamic adjustment of structural coordinates were both used to speed up the iterative convergence. Experimental results show that the algorithm can both improve image quality and accelerate the synthesis process. The speed up ratio varied due to texture randomness. For most structured textures, it is at least 100 times faster.