Supervised Gabor-Based Kernel Locality Preserving Projections for Face Recognition

Locality Preserving Projections (LPP) is an unsupervised method which seeks to optimally preserve the neighborhood structure of the dataset. LPP has been used wildly, but it has limits to solve the classification problem, such as the ignorance of the label information. Supervised Kernel Locality Preserving Projections (SKLPP) can preserve within-class geometric structures and represent the complex nonlinear variations of the face manifold by nonlinear kernel projection. Kernel method projects data from low-dimensional space to high-dimensional space. It can overcome the difficult when it is hard to use linear method in low-dimensional. In this paper, a novel Supervised Gabor-based Kernel Locality Preserving Projections (SGKLPP) method was proposed. This method integrates the Gabor wavelet representation of face images and the Supervised Kernel Locality Preserving Projections methods and it is robust to variations of illumination and facial expression. Experiments are performed to test the proposed algorithm on ORL dataset and Yale dataset. Results show that our new algorithm outperforms Supervised LPP (SLPP) method, SKLPP method and Supervised Gabor-based LPP (SGLPP) method.