摘要
针对基于字典学习算法的计算效率低,且大多局限于处理单帧图像的问题,提出了一种基于亚像素块匹配和字典学习的超分辨算法,以实现对多帧图像的重构。采用亚像素块匹配方法对图像进行配准,依据配准结果构造低分辨率字典,并通过计算辅助图像块与目标图像块的相似度来选择用于重构的图像块。在Matlab平台上,将该算法用于静态图像和视频图像处理,获得了较好的重构效果。
For algorithms based on dictionary learning have image reconstruction, a super resolution algorithm based on low computational efficiency and are limited to single frame sub-pixel block matching and dictionary learning was intro- duced to realize the reconstruction of multiple-frame images. Firstly, the image registration is realized by sub-pixel block matching, then the low-resolution dictionary is constructed according to the result of registration to reduce the amount of calculation, and the patches for reconstruction are selected by calculating the degree of similarity between object im- age and auxiliary image. The algorithm is proved to achieve better reconstruction when used in still and motion image on MATLAB platform.
作者
徐煜明
宋佳伟
肖贤建
XU Yu-ming SONG Jia-wei XIAO Xian-jian(School of Information and Engineering, Changzhou Institute of Technology, Changzhou 213002, China College of Computer and Information, Hohai University, Nanjing 211100, China)
出处
《计算机科学》
CSCD
北大核心
2016年第8期304-308,共5页
Computer Science
基金
国家自然科学基金:基于仿生视觉感知机理的金属板带表面缺陷在线检测方法研究(61273170)资助
关键词
超分辨率
稀疏编码
亚像素块匹配
字典学习
Super resolution, Sparse coding, Sub-pixel block matching, Dictionary learning
