A New Sampling-Based SVM for Face Recognition

Abstract

Support vector machine (SVM) needs huge computation for large scale learning tasks. Sample selection is a feasible strategy to overcome the problem. From the geometry of SVM, it is clear that a SVM problem can be converted to a problem of computing the nearest points between two convex hulls. The convex hulls virtually determine the separating plane of SVM. Since a convex hull of a set only can be constructed by boundary samples of the convex hull, using boundary samples of each class to train SVM will be equivalent to using all training samples to train the classifier. In order to select boundary samples, this paper introduces a novel sample selection strategy named Kernel Subclass Convex Hull (KSCH) sample selection strategy, which iteratively select boundary samples of each class convex hull in high dimensional space (induced by kernel trick). Experimental results on face databases show that our KSCH sample selection method can select fewer high quality samples to maintain SVM with high recognition accuracy and quickly executing speed.