基于改进样本块匹配准则的图像修复  

Image Inpainting Based on Improved Sample Patch Matching Principle

作　　者：张君 常霞[1] 王利娟 ZHANG Jun;CHANG Xia;WANG Lijuan

机构地区：[1]北方民族大学<宁夏智能信息与大数据处理重点实验室>

出　　处：《科技视界》2020年第2期15-18,共4页Science & Technology Vision

基　　金：国家自然科学基金项目（61561001）;宁夏自然科学基金项目（2018AAC03126）;宁夏高等学校一流学科建设（数学学科）（NXYLXK2017B09）;北方民族大学重大专项项目（ZDZX201801）资助;北方民族大学研究生创新项目（YCX19132）

摘　　要：基于样本块的图像修复方法是图像修复中常用的方法,有修复速度快,处理效果好,尤其可以处理大面积破损区域的图像等优点。基于样本块的图像修复方法在进行修复时,置信度项逐渐衰减为零,使得优先权计算的准确性明显降低,导致后期样本块的修复顺序发生较大的误差,产生图像局部模糊等情况。针对以上问题,本文引入引导因子和改用L1距离作为新的匹配准则。根据实验处理结果,本文改进的图像修复方法具有更佳的视觉效果和更强的鲁棒性,缩短了修复所用时间,取得了较好的修复效果。Image inpainting method based on sample patch is a common method in image inpainting.It has the advantages of fast repair speed and good processing effect,especially the image of large damaged area.When the image inpainting method based on the sample patch is carried out,the confidence term gradually decays to zero,which significantly reduces the accuracy of priority calculation,leading to large errors in the repair sequence of the sample patch in the later stage,resulting in local blur of the image,etc.In view of the above problems,this paper introduces the guide factor and uses L1 distance as a new matching criterion.According to the experimental results,the improved image restoration method in this paper has a better visual effect and stronger robustness,shortens the repair time,and achieves a better repair effect.

关 键 词：图像修复 样本块 优先权 置信度项 匹配准则 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]