Combining specific learning and generic learning for single-sample face recognition

Abstract

Single Sample Per Person (SSPP) problem, which means that there is only one training sample for each gallery subject, is a great challenge for face recognition to date. In this work, we address this problem by presenting a novel framework which combines specific learning and generic learning. The proposed approach is directly inspired from the complementarity between specific learning and generic learning. The former takes full advantage of the gallery samples and attempts to seek a low-dimensional subspace which can maximize the class separability, while the latter is able to provide complementary discriminative information by resorting to an auxiliary generic dataset with multiple samples per person. Experiments on FERET face dataset demonstrate the superiority of the proposed framework.