Determining optimally orthogonal discriminant vectors in DCT domain for multiscale-based face recognition
Yanmin Niu, Xuchu Wang · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2011
This paper presents a new face recognition method that extracts multiple discriminant features based on multiscale image enhancement technique and kernel-based orthogonal feature extraction improvements with several interesting characteristics. First, it can extract more discriminative multiscale face feature than traditional pixel-based or Gabor-based feature. Second, it can effectively deal with the small sample size problem as well as feature correlation problem by using eigenvalue decomposition on scatter matrices. Finally, the extractor handles nonlinearity efficiently by using kernel trick. Multiple recognition experiments on open face data set with comparison to several related methods show the effectiveness and superiority of the proposed method.