图约束字典和加权稀疏表示人脸超分辨率算法  

Face Hallucination Using Graph Regularized Dictionary and Weighted Sparse Representation

作　　者：黄克斌[1,2] 胡瑞敏[2] 王锋[1] 韩镇[2] 卢涛[2] 江俊君[2] 黄克斌 王锋[3] 

机构地区：[1]黄冈师范学院数字媒体技术系,湖北黄冈438000 [2]武汉大学国家多媒体软件工程技术研究中心,湖北武汉430072 [3] 黄冈师范学院数字媒体技术系,湖北黄冈,438000

出　　处：《电视技术》2014年第9期46-49,110,共5页Video Engineering

基　　金：国家"973"计划项目(2009CB320906);国家自然科学基金项目(60970160;61070080;61003184;61172173)

摘　　要：针对低分辨率、低质量人脸图像重建问题,提出了一种新的基于稀疏表示的人脸超分辨率算法。在训练阶段,人脸的位置特征被用于保持人脸块的全局信息,人脸块间的几何结构被用于保持高低分辨率超完备冗余字典的流形结构,从而提高字典的表达能力;在重建阶段,K近邻加权稀疏表示被用于消除稀疏编码噪声,以提高高分辨率人脸图像重建系数的精度。实验结果表明,提出的方法取得了较好的主客观质量。In this paper,a novel sparse representation based face hallucination method is proposed to reconstruct a high resolution( HR) face image from a low resolution( LR) observation. In training stage,it aims to get a more expressive HR-LR dictionary pair for certain input LR patch separately. Position feature of aligned face images is used to keep global similarity information of face image. The intrinsic geometric structure of training samples is incorporated into the spare coding procedure of the dictionary learning,which improve the expressive of the dictionary. In reconstruction stage,the weighted sparse representation is used to find an optimal coefficient for HR face image patch reconstruction. Experimental results validate the proposed method.

关 键 词：图约束字典 加权稀疏表示 人脸超分辨率 

分 类 号：TN919.18[电子电信—通信与信息系统]