基于深度学习的图像超分辨率复原研究进展  

Review on Deep Learning Based Image Super-resolution Restoration Algorithms

作　　者：孙旭[1] 李晓光[1] 李嘉锋[1] 卓力[1] 

机构地区：[1]北京工业大学信号与信息处理研究室,北京100124

出　　处：《自动化学报》2017年第5期697-709,共13页Acta Automatica Sinica

基　　金：国家自然科学基金(61471013;61370189;61372149;61531006);北京市自然科学基金(4142009;4163071);北京市属高等学校高层次人才引进与培养计划(CIT&TCD201404043;CIT&TCD20150311);北京市教育委员会科技发展计划(KM201510005004;KM201410005002);北京市属高等学校人才强教计划(PHR(IHLB))资助~~

摘　　要：图像超分辨率复原(Super resolution restoration,SR)技术是图像处理领域的研究热点,在视频监控、图像处理、刑侦分析等领域具有广泛的应用需求.近年来,深度学习在多媒体处理领域迅猛发展,基于深度学习的图像超分辨率复原技术已逐渐成为主流技术.本文主要对现有基于深度学习的图像超分辨率复原工作进行综述.从网络类型、网络结构、训练方法等方面分析现有技术的优势与不足,对其发展脉络进行梳理.在此基础上,本文进一步指出了基于深度学习的图像超分辨率复原技术的未来发展方向.Super resolution image restoration technology is a hot field of image processing in the field of video surveillance, image processing, forensic analysis, with a wide range of application requirements. In recent years, the rapid development of deep learning in the field of multimedia processing, deep learning based super-resolution images restoration has gradually become a mainstream technology. This paper reviews the existing deep learning based image super-resolution restoration work. In terms of network type, network structure, and training methods, the advantages and disadvantages of the prior art are analyzed and the development contexts are sorted out. On this basis, the paper further points out the future direction of the restoration technique based on deep learning of the super-resolution image.

关 键 词：超分辨率复原 深度神经网络 卷积神经网络 循环神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]