基于非局部相似字典学习的人脸超分辨率与识别  

Non-local Similarity Dictionary Learning Based Super-resolution for Improved Face Recognition

作　　者：廖海斌[1,2] 陈友斌 陈庆虎[3] LIAO Haibin CHEN Youbin CHEN Qinghu(School of Computer Science and Technology, Hubei University of Science and Technology, Xianning437100, China Guangdong Micropattern Software Co. , Ltd. , Dongwan 523808, China Electronic Information School , Wuhan University, Wuhan 430072, China)

机构地区：[1]湖北科技学院计算机科学与技术学院,湖北咸宁437100 [2]广东微模式软件股份有限公司,广东东莞523808 [3]武汉大学电子信息学院,湖北武汉430072

出　　处：《武汉大学学报（信息科学版）》2016年第10期1414-1420,共7页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金(61271256);国家公安部应用创新项目(2008YYCXHBST068);中国博士后科学研究面上支助项目(2015M582355);湖北科技学院博士启动基金项目(BK1418)~~

摘　　要：随着视频监控应用的普及,超低分辨率人脸识别问题越来越突出。现存的人脸识别算法在面对超低分辨率人脸图像时无法给出满意识别性能。在一定程度上,人脸超分辨率方法可以提高人脸的分辨率,但是,目前主流的基于字典学习的人脸超分辨率方法并不能很好地处理超低分辨人脸图像重建问题,尤其是超分辨率人脸识别问题。利用人脸图像块的非局部相似性和多尺度相似性,提出一种改进的基于字典学习的超分辨率人脸重建算法,同时提出尺度不变特征的超低分辨率人脸识别方法。实验结果表明:本文提出的方法不但具有很好的视觉效果,而且还具有很好的识别效果,与目前主流的人脸超分辨率和识别算法相比具有明显的优势。The Very Low Resolution（VLR）problem happens in many face recognition application systems given the increasing demand for camera-based surveillance applications,.Currently,the existing face recognition algorithms cannot deliver satisfactory performance with VLR face images.While face super-resolution（SR）methods can be employed to enhance the resolution of the images,the existing dictionary learning-based face SR methods are inadequate for VLR face images.To overcome this problem,we propose a novel SR face reconstruction method based on non-local similarities and multiscale linear combinations and subsequently,a new approach for VLR face recognition based on resolution scale invariant features.Experimental results show that the proposed approach based on dictionary learning outperforms the existing algorithms in public face databases,obtaining agood visuality suitable for face recognition applications subject to the VLR problem.

关 键 词：人脸超分辨率 人脸识别 字典学习 线性组合 非局部相似 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]