基于腐蚀和分块处理的Criminisi修复算法  

Criminisi Inpainting Algorithm Based on Corrosion and Block Processing

作　　者：刘正男 梁美玉 LIU Zhengnan;LIANG Meiyu(School of Electrical and Electronic Engineering,Anhui Institute of Information Technology,Wuhu Anhui 241000,China)

机构地区：[1]安徽信息工程学院电气与电子工程学院,安徽芜湖241000

出　　处：《海南热带海洋学院学报》2021年第2期73-80,共8页Journal of Hainan Tropical Ocean University

基　　金：安徽省质量工程新工科研究与实践项目(2018jyxm0321)。

摘　　要：由于传统Criminisi算法缺乏灵活性,且对于区域较复杂的图像,该算法只能够处理简单的相似图像块的复制填充等问题,据此提出一种改进的图像修复方法。该改进算法首先需要对掩膜图像进行腐蚀操作,避免人为造成的修复面积过大;然后针对优先权函数在修复后期不起作用的情况,引入信息熵进行了优化;最后针对Criminisi算法的相似块填充部分,提出进行分块修复,解决了修复复杂部分所造成的恢复混乱的情况。实验结果证明:改进算法取得了比较好的视觉效果,峰值信噪比高出原算法3.75 dB。该算法有效地进行了目标对象细节信息的恢复,图像质量得到了进一步提升。In order to solve the shortcomings of the traditional Criminisi algorithm,such as lacking of flexibility and limited capability of copying and filling in simple similar image blocks when dealing with image with a more complex regions,an improved image inpainting algorithm was proposed.Firstly,the improved algorithm corrodes the mask image to avoid the artificially large repair area.Secondly,to optimize the priority function in case it does not work in the later stage of the repair,the information entropy was introduced.Finally,as for the similar block filling part of Criminisi algorithm,a block repair was proposed to solve the recovery confusion caused by repairing the complex part.Experimental results proved that the proposed algorithm has better visual effects,with the peak signal-to-noise ratio 2.68 dB,higher than that of the original algorithm.Moreover,the present algorithm effectively restores the details of the image and improves the image quality.

关 键 词：图像修复 Criminisi算法 腐蚀 信息熵 分块处理 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]