Semi-supervised dimensionality reduction based on kernel marginal fisher analysis and sparsity preserving

Considering the limit that marginal fisher analysis(MFA) can´t take advantage of the discriminant information in the training samples, this paper proposed a semi-supervised dimensionality reduction based on kernel marginal fisher analysis and sparsity preserving. The new algorithm firstly gets the sparse reconstruction of the samples. Secondly it uses the samples with labels to construct the intra-class `similarity´ graph and inter-class `penalty´ graph. Then the algorithm uses all of the samples to get the global information. At last, we make it nonlinearized. The algorithm takes advantage of the information in both the label samples and unlabel samples. Experiments with the proposed algorithm were conducted on YALE and ORL, our algorithm outperforms based on traditional dimensionality reduction algorithms with maximum average recognition rate by 2.48% and 4.88% respectively.