Abstract: Locality preserving discriminant analysis takes very important position in face recognition research.Based on this,the 2DLPDA method was proposed,which directly processes the operation in the two dimensional space.In some way,it improves the performance of the algorithm.But the problem of the sensitivity to such variations like lighting’ expression and occlusion will make a big influence on the recognition rate when using 2DLPDA method.We proposed an improved algorithm called modular two-dimensional locality preserving discriminant analysis method.We divided the sample in blocks,so that we could extract the local neighborhood of the sample better.Because each sample was divided into blocks,the different blocks of one sample may have different nearest neighbors,causing local features of the sample to be extracted better .At the end of the method,all the local features are integrated together to be the basis for the identification.Experimental results on AR,YALE and ORL face databases show that the proposed method outperforms the 2DLPDA method.

Key words: Face recognition,Pattern recognition,Feature extraction,Locality preserving projection,Modular method,Maximum margin criterion