Constructing SVM multiple tree for face membership authentication

Abstract

In membership authentication problem that has a complicated and mixed data distribution, the authentication accuracy obtained from using one classifier is not sufficient despite its powerful classification ability. To overcome this limitation, an support vector machine (SVM) multiple tree is developed in this paper according to a "divide and conquer" strategy. It is demonstrated that the proposed method shows a good membership authentication performance, as well as the strong robustness to the variations of group membership, as compared with the SVM ensemble method [1]. Specifically, the proposed method shows a better improvement in authentication performance as the group size increases larger. © Springer-Verlag Berlin Heidelberg 2004.