基于稀疏表征多分类器融合的遮挡人脸识别.pdfVIP

Vol. 30 No.6 计算机应用研究 第 30 卷第 6 期 2013 年6 月 Application Research of Computer百 Jun. 2013 基于稀疏表征多分类器融合的遮挡人脸识别* 邓楠1 ，徐正光1 ，王瑭2 (1.北京科技大学控制科学与工程学院，北京 1∞083; 2. 西北工业大学电子信息学院，西安 71∞72) 摘 要:为了同时利用人脸局部信息，提出一种基于稀疏表征多分类器融合的遮挡人脸识别方法。先对人脸进 行多分辨率分块，求取并根据各子块稀疏表征分类器的识别率确定其权重，计算其后验概率估值，最终利用加权 融合准则进行多分类器融合识别。在 AR 和 YaleA 库的实验结采表明，该算法结果比稀疏表征遮挡人脸识别的 效果更好，鲁棒性更高。 关键词:人脸识别;稀疏表征;多分辨率分块;多分类器融合;过完备字典 文章编号: 1∞1-3695(2013)06-1914-03 中图分类号:T归91 文献标志码:A doi: 10. 3969/j. issn. 1001-3695.2013.06.083 Sparse representation based multi -classifier fusion approach for occluded face recognition 2 DENG Nan , XU Zheng-guang , WANG Jun ( 1. Institute of Control Science Engineering , Un阴阳ty of Sc啕lCe Technology Beijing , Beijing 1仪lO83 ， China; 2. Sch∞1 of Electronics I呐m创酬， North阳tem Pol附chnical University , Xi an 71∞72 ， Ch阳) Abstract; 币lÎs paper proposed a sparse representation based multi-classification fusion (WMSRC) algorithm on the basis of analyzing different discrimination ability of the various classifiers. In WMSRC , it firstly divided the face image into partitions by multi-resolution blocking. Then for each block ,it obtained SRC based classifiers. Finally ,performed the multiple classifiers fusion by weighted fusion criterion , which based on the weight and posterior probability of the various classifiers. Experiment results on the AR and Yale database prove that the algorithm performance is efficient. Key w