Face Recognition using Laplacian+OPRA-faces
Xiangyun Qing, Xingyu Wang · 2006
We propose a face recognition method named the Laplacian+OPRA-faces approach based on the Laplacian-faces and the OPRA-faces approach. An explicit mapping from the high-dimensional data space to the reduced space is obtained. The method aims to preserve both the local structure of Local Preserving Projections (LLE) and the local geometry of Orthogonal Projection Reduction by Affinity (OPRA). Several improved measures and selection method of parameters in the supervised case are presented. We compare the proposed approach with eigenfaces, fisherfaces, Laplacianfaces and OPRA-faces on two different face data sets. Experimental results indicate that Laplacian+OPRA-faces are better than Laplacianfaces and OPRA-faces.