Classifier Fusion Algorithm in Face Recognition

With the help of two-dimensional independent component analysis (2DICA) based on wavelet-transform (WT-2DICA), high-order statistical information can be extracted effectively, and experimental results on ORL (Olivetti Research Laboratory) and Yale face database show that correct recognition rate by WT-2DICA is good. However, this method is not valid to damaged images. Category information by improved Fisher can be considered sufficiently, whereas small samples problem by traditional Fisher can be avoided. Combined with the advantages of improved Fisher and WT-2DICA, fusion projection direction is obtained in this paper, which integrates best projection direction of improved Fisher and independent basis subspace of WT-2DICA. Experimental results on ORL and Yale face database show that the fusion method in this paper is valid in face recognition, and robust on damaged database