Abstract: The parameter-free locality preserving projection (PFLPP) is an effective feature extraction algorithm for face recognition, but it cannot effectively determine the neighbor relationship because it does not consider neighborhood relationship between the samples from different classes, and this algorithm judges the neighborhood relationship only by the distance between the samples and the population mean. In this paper, parameter-free uncorrelated maximum discriminant margin algorithm is proposed, which effectively uses the class information of the samples and needn’t set any parameters. The algorithm defines the similar weights of the neighbor samples from the same class and the punishment weights of the neighbor samples from different classes. The size of the sample neighborhood can be adaptively determined by the mean of the intraclass cosine distance and the inter-class cosine distance. In order to further enhance the performance of the algorithm, the uncorrelated objective function based on the maximum discriminant margin is put forward. The experimental results of UMIST and AR face database show that the proposed method has the advantages of low computation and high recognition performance compared with PFLPP and uncorrelated locality preserving projections analysis.

Key words: face recognition, feature extraction algorithm, parameter-free, uncorrelated