一种基于圆形区块随机增长的多样图约束纹理合成算法

A Constrained Multi-sample Texture Synthesis Algorithm Based on Random Expanding of Circular Patches

本文提出了一种基于圆形区块随机增长的多样图纹理合成算法改善了扫描线算法所带来的锯齿效应.算法利用高斯分布约束各个输入样图在合成结果中的比例,采用基于梯度引导的泊松平滑处理相邻纹理块之间的过渡区域,并在合成匹配函数中引入结构特征约束.实验结果表明,算法在引入比例约束的同时有效地保持了视觉效果和纹理结构的连续性.

This paper uses random expanding of circular patches to avoid the zigzag problem of current scan line multi-sample synthesis algorithm. Gaussian distribution is introduced to constrain the texture distribution ratio, and Poisson image editing is used to smooth the overlapped area of different textures. We also modify the match flmction to take the structural features into account. The experiment shows this new method can control the texture distribution while keeping the continuity of visual concept and texture structure.