Abstract: On the basis of the analysis of traditional elastic graph matching, a face recognition algorithm based on local feature analysis and optimization matching is proposed. Firstly, some important face features are locat ed by means of neutral network. Secondly, the multiscale features of the feature points are extracted with the local mutiscale analysis feature of the Gabor wavelet. In this way, every face feature point is represented by a series of Gabor wavelet coefficients. Finally, in order to find the face needed, the test face is compared with the multiscale features of the corresponding feature points in the face database with the optimization matching. Here the optimization matching method is proved strictly. The test results about Yale and ORL face database show that not only the proposed method is far better than the traditional EigenFace method but also the effect of the illumination variation on the face recognition is obviously overcome and the method has quite good robust for face expression variation in some degree.

Key words: local feature analysis, optimization matching, human face recognition, multiscale feature, Gabor wavelet