Face identification from a single example image based on Face-Specific Subspace (FSS)

In this paper, a novel face representation, named Face-Specific Subspaces (FSS), is proposed first. Then the corresponding face identification strategy is provided. This method is motivated, but essentially different, from Eigenface. In the proposed method, each face holds one private face subspace, while in Eigenface all faces share one common face subspace. To enable the proposed approach to tackle single example problem, a technique to derive multi-samples from single example is further developed. Extensive experiments on Yale face database Bern face database, and our 350 subjects database show that our method makes impressive performance improvement when compared with conventional Eigenface and template matching, which intensively indicates the robustness of our approach against appearance variances due to expression, illumination and pose.