Collaborative representation-based fuzzy discriminant analysis for Face recognition

In face recognition, the dimensionality reduction (DR) method is usually used to extract the discriminative features of the image. However, the performance is easily affected by varying facial poses, expressions and illumination. To solve this problem, a novel DR algorithm, namely collaborative representation-based fuzzy discriminant analysis (CRFDA), is proposed in this paper. In CRFDA, each training sample is firstly collaboratively represented by the overall training samples, and the fuzzy membership degrees of each sample are computed in terms of the representation coefficients. Secondly, the fuzzy means of different classes are computed using the membership degrees. Thirdly, the between-class and within-class scatter matrices are calculated to model the separability and compactness of samples, respectively. Finally, the feature extraction standard is improved by maximizing the ratio of fuzzy between-class scatter to fuzzy within-class scatter. A large number of experiments on publicly available facial datasets demonstrate the effectiveness of the proposed method.