关键词: 模式识别, 人脸识别, 协作表示, 模糊渐进构造, 图嵌入

Abstract: In the procedure of the construction of neighborhood graph, traditional graph-embedding algorithms adopt a simple two-value hard classifier criterion. Concerning this problem, with reference to the fuzzy mathematics, a new fuzzy progressive neighbor graph was proposed in this paper. Furthermore, collaborative representation classifies patterns by employing all the training images to represent the query image collaboratively. Therefore, in this paper, collaborative representation was introduced into classifier. Concerning the problems mentioned above, a feature extraction algorithm based on collaborative representation and fuzzy progressive maximal marginal embedding was proposed for face recognition. The experimental results on the ORL, AR face databases and USPS handwriting number database show that the proposed algorithm outperforms Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Localities Preserving Projections (LPP) and Margin Fisher Analysis (MFA).

Key words: pattern recognition, face recognition, collaborative representation, fuzzy progressive constructive, graph-embedding