快速非局部均值形态成分分析唐卡图像修复算法  

Fast Morphological Component Analysis Tangka Image Inpainting Based on Nonlocal Mean Filter

作　　者：胡文瑾[1,2] 李战明[1] 刘仲民[1] 

机构地区：[1]兰州理工大学电气工程与信息工程学院,兰州730050 [2]西北民族大学数学与计算机科学学院,兰州730030

出　　处：《计算机辅助设计与图形学学报》2014年第7期1067-1074,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61162021);甘肃省科技计划资助项目(096RJZA112);中央高校基本科研业务费专项资金资助(zyz2012078);西北民族大学创新团队

摘　　要：唐卡图像包含内容丰富、颜色复杂,使用基于形态成分分析(MCA)的图像修复算法修复图像时,增加全变分对恢复有毛糙边缘的分段光滑图像效果较好,但其很难有效利用图像全局冗余相似信息,并且易产生阶梯效应.针对该问题,提出一种快速的非局部均值MCA唐卡图像修复算法.该算法利用像素周围固定大小窗口内的信息表征该像素的特征,使得像素估计结果能够很好地保留结构细节信息,即分解后的结构部分具有更稀疏的梯度,使得后续的迭代分离效果更加优良;同时,使用该算法有效地减少了区域内不相关像素权值的计算,降低了算法的复杂度.对于唐卡图像中出现的折痕、斑块状破损以及部分信息缺失的修复实验结果表明,文中算法具有良好的修复能力.Thangka image contains rich content and complex color. In the context of image inpainting based on morphological component analysis, the imposition of a total variation penalty is useful, particularly well in recovering piecewise smooth objects, but global redundancy similar information is difficult to utilize effectively and easy to produce staircase. This paper proposed a new image inpainting method based on morphological component analysis. It utilized the information of pixels around fixed window to express the current pixel in order to preserve the fine structure and details and eliminated staircase simultaneously, which made the subsequent iteration more effective. In addition, the improved algorithm based on fast algorithm reduced the calculation weights of uncorrelated pixels within an area, thus decreased complexity of the algorithm. Experimental results for thangka image which contains scratch and block loss show that the proposed method achieves better inpainting effect.

关 键 词：图像修复 形态成分分析 全局信息冗余性 非局部均值 快速计算 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]