融合学习算法的单帧图像超分辨率复原  

Single image super-resolution combined with learning algorithm

作　　者：黄全亮[1] 刘水清[1] 孙金海 陈柯[1] 

机构地区：[1]华中科技大学电子与信息工程系,武汉430074 [2]电磁散射重点实验室,北京100854

出　　处：《计算机工程与应用》2013年第23期186-190,共5页Computer Engineering and Applications

基　　金：国家自然科学基金(No.60705018)

摘　　要：为从高度降质的单帧图像中重建出高分辨率图像,提出了一个结合自适应正则化与学习算法的超分辨率复原方法。该方法基于图像的局部特征,实现了正则化方法动态自适应控制过程,优化了学习算法中的训练集、预测原则和搜索过程,以降低基准图相关性要求、提高搜索效率。仿真实验分步论证了该方法的有效性,以及对复原效果的提升。In order to reconstruct a high-resolution image from single highly blurred image, a super-resolution reconstruction method combined with adaptive regularization and learning algorithm is proposed. Based on local characteristics, the dynamic adaptive control progress of reconstruction method is achieved. Train set, prediction principle and searching progress of learning algorithm is optimized to depress the relativity of example image and improve the searching efficiency. Experimental results demonstrate the availability of the method by steps, and the improvement of reconstruction result.

关 键 词：图像复原 超分辨率 正则化 学习算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]