基于演化算法的纹理合成图像修复技术

摘要

Aiming at some deficiencies in calculating the priority of the damaged patch and the inefficiency in searching optimal matching patches the Criminisi algorithm has, we propose a texture synthesis image inpainting technique based on evolutionary algorithms. The algorithm takes into account the isophote and gradient directions both simultaneously for the priority of the patch to be inpainted first, and controls the size of template window according to gradient change; then it uses evolutionary algorithm to search the optimal patch around the damaged region, and finally reaches the goal of restoring the image fast and accurately. Simulation experiments show that the proposed new algorithm performs better in convergence speed and inpainting effect compared with conventional algorithms, its comprehensive performance is superior to the Criminisi algorithm.%针对Criminisi算法中计算待修复块优先级时存在的一些不足及搜索最佳匹配块效率低的问题,提出一种基于演化算法EA(Evolutionary Algorithm)的纹理合成图像修复技术.该算法首先从等照度线方向和梯度方向共同考虑待修复块的优先权,并根据梯度的变化来控制模板窗口的大小,然后提出用演化算法在待修复区域周围搜索最佳匹配块,最终达到快速准确修复图像的目的.仿真实验表明,提出的新算法与传统算法相比具有更快地收敛速度和更好地修复效果,其综合性能优于Criminisi算法.