摘要
文物图像包含了丰富的内容,使用传统偏微分方法很难有效地恢复文物图像。为了有效恢复原图像,提出一种基于非局部矩阵填充的文物修复方法。该算法充分利用图像的冗余信息,在搜索窗口内,通过比较像素点周围一个矩形邻域内的多个像素点的相似性,找到一定数量的相似块转换为列向量,构成相似块组矩阵。由于相似块组矩阵为低秩或者近似低秩的,再通过矩阵填充的原理对损坏像素点进行填充修复。对于文物图像中出现的不同比例像素点丢失、有划痕、有文字破损的修复实验结果表明,该算法能够很好地保留结构细节信息,拥有较好的视觉特性,使目标轮廓边缘特征清晰,有效地恢复出文物图像的原貌,具有良好的修复能力。
Cultural relic images contain rich contents,and they are difficultly to be restored effectively with traditional partial differential methods. In order to effectively restore original images,we propose a non-local matrix completion-based method for cultural relic image restoration. The method makes full use of the redundant information of image,within the search window and by comparing the similarity of multiple pixels in a rectangular neighbourhood around the pixel,it finds a certain number of similar blocks and converts them to column vector so as to construct the similar block group matrix. Since the obtained similar block group matrix is low rank or near to low rank,we then use matrix completion theory to fill and restore the damaged pixels. It is demonstrated by the restoring experimental results covering the pixels missing in different proportions,scratches and words damages occurred in cultural relic images that the proposed method is able to well preserve the detail information of structure,has good visual characteristics and clear edge features of the target profile. It can effectively recover the original appearance of the cultural relic image with good restoring capability.
出处
《计算机应用与软件》
CSCD
2016年第11期126-129,184,共5页
Computer Applications and Software
基金
国家自然科学基金项目(51365017
61305019)
江西省科技厅青年科学基金项目(20132bab211032)
关键词
矩阵填充
非局部
相似块
低秩
Matrix completion
Non-local
Similar block
Low rank