基于边缘特征和像素结构相似度的图像修复算法

针对Criminisi算法的块匹配准则仅采取单一颜色判断因子导致无法合理选择最佳样本块,且其在修复过程中使用单一修复模板易出现填充裂纹和错误像素的问题,提出基于边缘特征和像素结构相似度的图像修复算法.首先提出一种局部特征与边缘纹理分辨相结合的分段修复算法以增强边缘纹理分辨能力;其次采用样本相似度和信息熵相似度确定最佳样本块集合,并依据颜色和特征项的欧氏几何距离及结构相似性确立块匹配准则;再通过基于信息熵的自适应修复模板解决Criminisi算法的填充裂纹和错误像素问题;最后引入果蝇优化算法以减少图像修复时间.实验结果证明,对于不同的图像,文中算法能取得较为满意的修复效果和修复效率.

Image Inpainting Algorithm Based on Edge Feature and Pixel Structure Similarity

The patch matching criterion of the Criminisi algorithm could not choose the best sample patch reasonably because single color factor was adopted only, and the single inpainting template could result in the filling cracks and the erroneous pixel during the inpainting process. A new algorithm was proposed to solve these problems. Firstly, a piecewise inpainting combining local features and edge texture resolutions was proposed to enhance the edge texture resolution. Secondly, the sample similarity and the information entropy similarity were used to determine the best sample patch set, and the patch matching criteria was established according to the texture similarity and the Euclidean geometry distance of the color and the feature items. Then, the filling cracks and the erroneous pixel problem of the Criminisi algorithm were solved by the adaptive inpainting template algorithm based on the information entropy. Finally, the fruit fly optimization algorithm was introduced to reduce the time of inpainting image. The experimental results showed that this new algorithm could achieve the satisfactory inpainting effect and the inpainting efficiency for different images.