Face Recognition Method Based on Improved LDA

An Improved LDA method is proposed in this paper. A weakness of Existing fuzzy LDA model is that the class mean vector in the expression of between-class scatter matrix is estimated by the class sample average while the mean vector in the definition of the within class scatter matrix is the fuzzy average of class samples. Under the non-ideal conditions, such as the variations of expression, illumination, pose, and so on, there will be some outliers in the sample set, so the average vector of all samples is not sufficient to estimate the center of the training samples accurately. As a result, the recognition performance of the traditional fuzzy LDA model will decrease. To address this problem, also to render the fuzzy LDA model rather robust, the membership degree matrix of training samples is calculated firstly by the fuzzy K-nearest neighbor (FKNN), which is used to get the fuzzy means of each class and the average of fuzzy means, then we incorporate them into the definitions of the within-class scatter matrix and the between class scatter matrix, respectively. Experiment results conducted on the ORL, YALE and FERET face databases indicate the effectiveness of the proposed approach.