Subspace Algorithms for Face Verification
In real-life applications of face recognition the task of verification appears to be more important than classification. Usually we only have a limited collection of training images of some person and we want to decide if the acquired photograph is similar enough to them without using a separate set of negative samples. In this case it seems reasonable to use a subspace method based on a coordinate system constructed individually for the given class (person). This work presents two such methods: one based on SDF and the other inspired by Clafic algorithm. In the experimental section they are compared to the two-class SVM on the realistic data set taken from CMU-PIE database. The results confirm the advantages of subspace approach.
KeywordsFace Recognition Training Image Subspace Method False Acceptance Rate False Rejection Rate
Unable to display preview. Download preview PDF.
- 1.Phillips, P.J., Scruggs, W.T., O’Toole, A.J., Flynn, P.J., Bowyer, K.W., Schott, C.L., Sharpe, M.: FRVT 2006, and ICE 2006, Large-Scale Results. NISTIR 7408 National Institute of Standards and Technology, Gaithersburg (2007)Google Scholar
- 5.Watanabe, S., Lambert, P.F., Kulikowski, C.A., Buxton, J.L., Walker, R.: Evaluation and Selection of Variables in Pattern Recognition. Computer and Information Sciences II, 91–122 (1967)Google Scholar
- 6.Watanabe, S.: Karhunen-Loeve Expansion and Factor Analysis. In: Trans. Fourth Prague Conf. on Information Theory, Statistical Decision Functions, Random Processes, pp. 635–660 (1965)Google Scholar
- 8.Smiatacz, M., Malina, W.: Modifying the Input Data Structure for Fisher Classifier. In: Proc. Second Conf. on Computer Recognition Systems (KOSYR 2001), pp. 363–367 (2001)Google Scholar
- 10.Sim, T., Baker, S., Bsat, M.: The CMU Pose, Illumination, and Expression (PIE) Database. In: Proc. 5th Int. Conf. on Automatic Face and Gesture Recognition (2002)Google Scholar