Feature extraction using 2DIFDA with fuzzy membership

In this paper, a new method called fuzzy two-dimensional inverse Fisher discriminant analysis (fuzzy 2DIFDA) directly based on 2D image matrices rather than image vectors is proposed for feature extraction and recognition. In the proposed method, the distribution information of samples is first characterized using fuzzy set theory, and the corresponding fuzzy scatter matrices are then redefined. Image discriminant features which have embedded the fuzzy information are finally extracted by selecting 2D principal components and 2D inverse Fisher discriminant vectors. Experimental results on FERET face database and FKP database demonstrate the effectiveness of the proposed method.