联合Gabor误差字典和低秩表示的人脸识别算法.pdf

第 44 卷第3 期 计算机科学 Vo1. 44 NO.3 COMPUTER SCIENCE 岛1ar.2017 2017 年 3 月 联合 Gabor 误差字典和低秩表示的人脸识别算法 首照宇杨晓帆 (桂林电子科技大学认知无线电与信息处理教育部重点实验室 桂林 541004) 摘 要针对人脸图片的遮挡、伪装、光照及表情变化等问题，根据 Gabor 特征对遮挡、伪装、光照及表情变化有着更 强的鲁棒性的特点，提出了联合 Gabor 误差字典和低秩表示的人脸识别算法(GDLRR) 。首先对训练样本和测试样本 分别进行Gabor 特征提取，并将这些特征组成待测试的特征字典;然后将一个单位阵进行Gabor 特征提取并训练成一 个更紧凑的 Gabor 误差字典;最后联合 Gabor 误差字典和训练、特征字典对测试特征字典进行低秩表示后进行分类识 别。各类实验表明，提出的改进算法对人脸识别的各类问题都有着更强的鲁棒性和更高的识别准确率。 关键词 遮挡，低秩表示，Gabor 特征，误差字典，降维 中图法分类号 TP39 1. 41 文献标识码 A 001 10. 11896/j. issn. 1002-137叉 2017.03.060 Jointing Gabor Error Dictionary and Low Rank Representation for Face Reco伊ition SHOU Zhao-yu YANG Xiao-fan (Key Laboratory of Cognitive Radio and Infonnation Processing ,Guilin University of Electronic Technology ,Guilin 541004 ,China) Abstract Focused on the issues that face images have the problems occlusion ,disguise ,illumination and facial expres sion changes in face recognition ,an improved face recognition method was proposed. According to the characteristics of the Gabor feature for occlusion , disguise , illumination and facial expression changes has stronger robustness ,jointing Gabor error dictionary and low rank representation CGDLRR) for face recognition was proposed. Firstly ,the Gabor fea ture of training samples and testing samples are extracted for making up features dictionaries that is to be tested ,re spectively. And then ,a unit matrix is utilized to extract Gabor feature for training a more compact Gabor error dictiona ry. Finally ,lowest-rank representation of feature dictionary of testing samples is sought for classification b