中文摘要：

如果每类训练样本较充分,基于稀疏表示分类可以取得比较好的识别效果;当训练样本比较少时,它的分类效果可能就不理想。拓展的稀疏分类算法可以较好地解决这一问题,它在表示测试样本时,引入了训练样本的类内变量矩阵,利用它和训练样本集来表示测试样本,从而提高了人脸识别率。然而,该算法并没有考虑训练样本在表示测试样本中所起的作用,即所有训练样本的权重都等于1。采用高斯核距离对训练样本加权,提出用加权的训练样本和类内散度矩阵来共同表示测试样本,即基于加权的拓展识别算法。实验证明所提算法能够取得更好的人脸识别效果。

英文摘要：

The sparse representation-based classification（ SRC） can achieve the good recognition result if each class has sufficient training samples. The recognition result of SRC is not desirable if very few training samples per class are available. To address this problem,extended sparse representation-based classification（ ESRC） applied the auxiliary intraclass variant dictionary and the training sample set to represent the test sample and could improve the face recognition performance. However,the algorithm did not consider the contribution of training samples in the representation of test sample. This paper used the weighted training samples which were weighted by the Gaussian kernel distance and the within-class intraclass matrix to jointly represent the test sample,and proposed algorithm called weighted extended sparse representation for classification algorithm（ WESRC）. Experiments show that the proposed algorithm can achieve better classification results.