一种改进的在线字典学习图像超分辨率重建算法  

An improved super-resolution image reconstruction algorithm based on online dictionary learning

作　　者：刘冬雪[1] 秦晅[2] 杨欣[1] 周大可[1] LIU Dong-xue;QIN Xuan;YANG Xin;ZHOU Da-ke(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China;The 28 th Research Institute of China Electronics Technology Group Corporation,Nanjing 210007,China)

机构地区：[1]南京航空航天大学自动化学院,江苏南京210016 [2]中国电子科技集团公司第二十八研究所,江苏南京210007

出　　处：《云南民族大学学报（自然科学版）》2018年第2期140-146,共7页Journal of Yunnan Minzu University:Natural Sciences Edition

基　　金：国家自然科学基金(61573182)

摘　　要：为提高重建图像的质量,详细表达图像高频细节信息,提出了一种改进的在线字典学习图像超分辨率重建算法.该算法在稀疏重建的字典训练阶段,采用在线字典学习以获取最优的超完备字典.在稀疏系数表示阶段,考虑图像多尺度间的冗余信息,构造L1范数正则项补偿对,抑制稀疏系数噪声提高重建效果.实验表明,该算法可更好地恢复图像细节,在客观评价和主观视觉感知上图像的重建质量均有所提高.In order to improve the quality of the reconstructed image,and express the high frequency details of the image,an improved super-resolution image reconstruction algorithm for online dictionary learning is proposed.In the dictionary training stage of sparse reconstruction,the algorithm uses online dictionary learning to obtain the optimal super-complete dictionary.In the sparse representation stage,considering the redundant information between multi-scale images,it constructs L1 norm regularization compensation pairs,restrains sparse coefficient noise and improves reconstruction effects.Experiments show that the algorithm can better restore image details and improve the quality of image reconstruction in objective evaluation and subjective vision perception.

关 键 词：字典学习 稀疏表示 超分辨率重建 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]