A new algorithm called fuzzy inverse fisher disciminant analysis is proposed. The inverse fisher discriminant analysis (IFDA) is effective in extracting discriminant features

but is assumed the same level of relevance of each sample to the corresponding class. Distribution information of samples is represented with fuzzy membership degree in the fuzzy inverse fisher discriminant analysis (FIFDA). Furthermore the information is utilized to redefine the corresponding scatter matrices

which are different from the IFDA. The experimental results on ORL and FERET face database indicate that the performance of FIFDA is superior to that of Fisherfaces and IFDA.