A perturbation algorithm for MDA in high dimensional space and its application to face recognition
Wenming Zheng · Journal of Circuits and Systems · 2004
A perturbation algorithm for the multiple discriminant analysis (MDA) in high dimensional space is presented. In classical MDA method, scatter matrix within-class often may be singular, especially in face recognition. This usually comes from so-called “small sample size” (SSS) problem, i.e., the dimensionality of input space is higher than the number of all samples. Accordingly, a novel algorithm is proposed to tackle this problem. Experiment on ORL face database shows that compared with eigenface and Fisherface method, using the proposed method, the average MDA error rate is only 3.23%, which is 49.9% of the traditional eigenface and 79.4% of the Fisherfaces method.