基于同类测试样本组的稀疏表示人脸识别

Sparse Representation Classification for Face Recognition with Intra-class Testing-sample Group

近年来的研究表明,稀疏表示分类(SRC)方法是一种有效的人脸识别方法。SRC是单个样本基于向量l1-范数正则化的最小二乘分类。但现实中常常存在着已知多个测试样本属于同一类的情况,无疑有利于分类,而基于SRC或其他单样本模型的方法却未能利用该信息。为利用类别标签信息,提出了一种新的鲁棒人脸识别方法。该方法基于同类测试样本组的稀疏表示分类(IGSRC),即将同类多个测试样本放至同组,采用矩阵L1-范数正则化的最小二乘分类进行稀疏表示,将测试样本组判为类别中残差最小的标号。实验结果表明,相比于SRC与IGSRC方法,所提出的方法不但能取得更高的人脸识别率(即使在每类别训练样本数较少、训练样本存在部分遮挡),而且具有更少的计算耗时。

Recent studies have shown that Sparse Representation Classification （SRC） is an effective method for face recognition. SRC is a least squares classification based on l1 - norm regularized for a single testing-sample. However,in the case that multiple testing-sam- ples are known to be die same class which is surely helpful in the classification,the common-class label information is not included in SRC or other single-sample models. Therefore, a novel robust face recognition method based on sparse representation classification is pro- posed which is on the basis of IGSRC. Taking multiple intra-class testing-samples into the same group, it adopts the matrix L1 - norm regularized least squares classification for sparse representation and judges the test sample group as the label with minimum error in clas- ses. Experimental results show that compared with IRC and IGSRC, the method proposed cannot only obtain better face recognition rate （even when the number of training samples per subject is small or training samples are partly occluded） ,also own less running time.