Face Recognition Using Affine Subspace Nearest Points Approach

Abstract

This paper proposed a novel classification approach, called affine subspace nearest points (ASNP) approach, for face recognition. Inspired by the nearest point problem of SVM which computes the nearest points between two convex hulls, ASNP replaces the convex hulls with the affine subspaces of each class samples, i.e., ASNP approach constructs the two smallest subspaces which respectively contain the data in each class, and finds the closest points in two subspaces. Then, the plane, which bisects the line segment connecting the two points, is constructed as the separating plane. Compared with SVM, ASNP is simply a linear optimal problem, and it avoids the convex quadratic programming problem. We apply the ASNP approach on face recognition. The experiments on ORL face database, Yale face database and Harvard face database show that the performances of ASNP algorithms are competitive to 1-NN classifier and SVM classifier for face classification.