Letters: Fuzzy MSD based feature extraction method for face recognition

摘要: To improve the recognition performance of maximum scatter difference (MSD), a fuzzy MSD method is proposed in this paper. In existing model, class mean vector expressions within-class matrix and between-class estimated by sample average. Obviously, average not sufficient to provide an accurate estimate using few given samples, because there will be some outliers set under non-ideal conditions such as variations expression, illumination, pose, so on. As result, traditional model decrease. address problem, inspired application, theory incorporated into algorithm. method, applying K-nearest neighbor (FKNN), membership degree training samples calculated, which used get means each then applied definition between matrix, respectively. The results experiments conducted on ORL, YALE FERET face database indicate effectiveness approach.