Gabor Feature Based Classification Using LDA/QZ Algorithm for Face Recognition
This paper proposes a LDA/QZ algorithm and its combination of Gabor Filter-based features for the face recognition. The LDA/QZ algorithm follows the common “PCA+LDA” framework, but it has two significant virtues compared with previous algorithms: 1) In PCA step, LDA/QZ transforms the feature space into complete PCA space, so that all discriminatory information is preserved, and 2) In LDA step, the QZ-decomposition is applied to solve the generalized eigenvalue problem, so that LDA can be performed stably even when within-class scatter matrix is singular. Moreover, the Gabor Filter-based Features and the new LDA/QZ algorithm are combined for face recognition. We also performed comparative experimental studies of several state-of-art dimension reduction algorithms and their combinations of Gabor feature for face recognition. The evaluation is based on six experiments involving various types of face images from ORL, FERET, and AR database and experimental results show the LDA/QZ algorithm is always the best or comparable to the best in term of recognition accuracy.
KeywordsFace Recognition Linear Discriminant Analysis Face Image Generalize Eigenvalue Problem Discriminatory Information
Unable to display preview. Download preview PDF.
- 1.Zhao, W., Chellappa, R., Rosenfeld, A., Phillips, P.J.: Face Recognition: A Literature Survey. ACM Computing Surveys, 399–458 (2003)Google Scholar
- 6.Martinez, A.M., Benavente, R.: The AR Face Database (2003), http://rvl1.ecn.purdue.edu/aleix/aleix_face_DB.html