摘要
针对红外图像存在分辨率不高、对比度低的特点,提出了基于引导滤波和多尺度局部自相似性的红外单幅图像超分辨率算法。该方法引进了类高斯分布的类高斯核,在此基础上构建均值引导滤波器。该滤波器是一种线性边缘保持滤波器,可以得到图像的高频细节。根据图像的自相似性,对初始高分辨率图像和原始低分辨率图像进行分块,得到待匹配窗和搜索窗,根据非局部均值(NLM),待匹配窗图像块的值利用搜索窗中相似块的加权平均计算得到;利用图像自相似性,待匹配窗在搜索窗的邻域内进行匹配搜索,找到与待匹配窗最相似的匹配块,计算出最佳匹配块的高频细节图像块,与相似块的加权平均值相加,重构出高分辨率待匹配窗;最后,合并所有的超分辨率重构的待匹配窗,相邻图像块重叠区域的像素值使用平均融合得到,得到最终的超分辨率图像。实验结果表明,本算法不仅能很好地重构图像的高频细节,还能很好地恢复图像的纹理特征,得到的结果不仅边缘更清晰更真实,而且纹理更加丰富。
Aiming at the characteristics of low resolution and low contrast of infrared image,this paper proposed a new super resolution method from a single infrared image based on guided filtering and multi-scale local self-similarity.First of all,this method introduced a Gauss kernel,which was similar to the Gauss distribution.Based on this,it constructed the mean direct filter,the filter was a linear edge preserving filter,which could get the high frequency details of the image.Again,according to the self similarity of the image,it divided the initial high resolution image and low resolution of the original image into patchs,to obtain the matching window and the searching window,according to the non local mean(NLM),the value of matching window was the weighted mean value of searching window.Secondly,it used image’s self similarity,matching window searching and matching in the neighborhood search window,and found the matching window matching patch was the most similar,calculated the best matching image detail patchs,together with the weighted average value of similar blocks,to reconstruct high resolution matching window.Experimental results show that the proposed algorithm can not only reconstruct the high frequency details of the image,but also restore the texture features of the image.The results obtained are not only more clear and real,but also rich in texture.
作者
刘哲
黄世奇
姜杰
Liu Zhe;Huang Shiqi;Jiang Jie(College of Electronic&Information Engineering,Xijing University,Xi’an 710123,China)
出处
《计算机应用研究》
CSCD
北大核心
2018年第4期1236-1240,1245,共6页
Application Research of Computers
基金
国家自然科学基金资助项目(61379031)
校基金资助项目(XJ150222)
关键词
超分辨率
引导滤波
多尺度
局部自相似
super resolution
mean guided filter
multi-scale
self-similarity
