动态尺度块匹配约束下的链式优化图像修复研究

Image Completion Using Dynamic-Scale Patch Matching and Layer-Wise Chain Optimization

针对由块间像素值突变引起的修复质量下降问题,提出一种基于动态尺度块匹配的链式优化图像修复算法.该方法依据先验信息与结构特征确定当前修复层所有图像块的候选匹配块数量;构建多尺度块匹配搜索模型,以二次搜索定位目标候选块,形成图像修复可行解空间.在候选块集合约束下,建立基于块间关联匹配特性的链式优化修复模型,并采用动态规划法求解当前修复层的最优匹配块集合,实现对图像的由外向内修复.实验结果表明,算法对多种自然图像的修复达到了令人满意的效果.

In order to reduce visually inconsistent results caused by sudden change of pixel values between patches,a novel image completion method based on dynamic-scale patch matching and layer-wise optimization was proposed. During patch search- ing, the number of candidate patches for the current layer was calculated through the analysis of prior knowledge and structure fea- tures;meanwhile, a multi-scale patch searching model was given to obtain the best candidate patches. Those patches constituted the feasible solution space for image completion. With the intrinsic characteristics and relevance of adjacent patches taken into consider- ation,image completion was abstracted as a chain optimization problem. The layer-wise chain optimization model was established and solved through dynamic programming. The optimal patches for the current layer were obtained and the image was repaired from the outside to the inside layer by layer. Experimental results demonstrate both the effectiveness and efficiency of the proposed algorithm for various natural images.