关于线性映射的灰度级超分辨率人脸识别方法.pdfVIP

第6卷 第7期 482 中国科技论文在线 SCIENCEPAPER ONLINE 2011 年 7 月 基于线性映射的灰度级超分辨率人脸识别方法 张琳彦，吴 宁，齐 春 （西安交通大学电子与信息工程学院，西安 710049 ） 摘 要：针对现有低灰度级分辨率人脸识别算法的识别率不高、计算复杂度高的问题，提出了一种基于特征域线性映 射的灰度级超分辨率人脸识别算法。选择Principal Components Analysis (PCA)为人脸识别特征空间，根据流行学习的相 关理论，假定高、低灰度级人脸图像识别特征之间存在线性的映射关系，并通过Procrustes分析建立了从低灰度级到高 灰度级特征的线性映射，根据该线性映射，即可得到输入低灰度级分辨率人脸图像所对应的高灰度级分辨率识别特征， 并通过最近邻分类器进行识别。实验结果表明，同现有算法相比，该算法识别率最高、耗时最短，并且识别率对特征 的维度不敏感。 关键词：特征空间；低灰度级分辨率；人脸识别；Procrustes分析；主成分分析 中图分类号：TN 91.1 文献标志码：A 文章编号：1673－7180(2011)07 －0482 －7 Gray-scale super resolution for face recognition based on linear mapping Zhang Linyan ，Wu Ning ，Qi Chun (Department of Electronic and Information Engineering , Xi’an Jiaotong University , Xi’an 710049, China) Abstract: Previous work on face recognition in low gray-scale resolution had high computation complexity, but not very high recognition rate. To deal with this problem, a gray-scale super-resolution algorithm utilizing linear mapping in feature-domain for face recognition was proposed. This method chose Principal Components Analysis (PCA) feature as feature-domain of face recognition, and assumed that linear mapping existed between high and low gray-scale resolution face recognition features according to the theory of manifold learning. Procrustes analysis was utilized to establish mapping from low to high gray-scale resolution features. Based on the mapping, high gray-scale resolution facial images recognition features could be obtained when the low gray-scale resolution