中文摘要：

完备的稀疏表示方法近年来应用在人脸识别中并取得较好的结果，它可以仅利用样本的随机投影完成对测试样本的识别。在实际应用中，由于受光照、遮挡等因素的影响，测试样本并不能通过训练样本的线性组合得到很好的稀疏重构。本文提出了基于Metaface字典学习与核稀疏表示的人脸识别方法，借助核技巧，将数据样本和字典集映射到高维的未知空间，以解决特征的非线性相似问题。在核空间对数据样本进行稀疏重构，得到数据在核空间的一种简洁的稀疏表达方式从而提高识别率，而Metaface字典学习框架的引入可以得到更加精炼的字典，从全局上提高识别率。通过在ORL人脸库、Yale人脸库和AR人脸库的实验表明，同等情况下，本文提出的方法优于PCA，SVM，SRC等方法，进一步提高了人脸识别率，具有较好的应用价值。

英文摘要：

The sparse representation classification （SRC） has been successfully applied in human face recognition, and can achieve the classification of the test sample by using only random projections of sample images. However, due to the light and expression change in practical application, the test sample cannot achieve sparse representation by the linear combination of training samples. This paper proposes a face recognition method based on Metaface and kernel sparse representation, which utilizes the kernel trick to map the original data and Metaface set into kernel space such that the nonlinear similarity of the feature can be solved. Moreover, in the kernel space, the original data are reconstructed by sparse representation to get a concision expression, which integrates the Metaface learning framework to improve the recognition rate globally. The test results on ORL database, Yale database and AR database indicate that the proposed method can achieve a higher recognition ratio than other classical methods such as standard SRC, PCA, and SVM.