Abstract: Few study has been performed on the nonlinear feature selection in the field of sparse representation face recognition. A hierarchical framework for sparse representation face recognition is proposed. First it constructs the face overcompleted dictionary of wavelet low frequency feature and high frequency feature. During the recognition, low frequency feature is used to sparse representation face recognition firstly, and then the category fuzzy coefficient is computed and decided to output the category tag or go on the high frequency sparse representation face recognition. The experimental results show that the face recognition hierarchical framework improves the accuracy of recognition and the framework is robust to face image occlusion.

Key words: face recognition, sparse representation, hierachical framework, wavelet feature

摘要： 基于稀疏表示的人脸识别研究，非线性特征的选择研究较少。提出分层使用人脸图像的小波特征，进行稀疏表示人脸识别框架。框架首先对样本人脸进行小波变换，构造小波低频和小波高频过完备人脸字典；识别阶段首先使用人脸图像的小波低频特征进行稀疏表示，计算类别模糊稀疏，然后根据模糊系数输出类别标签或进行高频特征的稀疏表示与识别。实验结果表明，基于小波特征和稀疏表示的人脸识别分层框架提高了识别的准确率，且对遮挡很鲁棒。

关键词: 人脸识别, 稀疏表示, 分层框架, 小波特征