View synthesis from Infrared-visual fused 3D model for face recognition

Abstract

Recently, there has been an increasing interest in infrared (IR) face recognition. However, few works are reported on the face pose problem in IR images. Existing methods have to capture tens of images of each subject at various poses, resulting in a huge database and longer registration time. This paper presents a method for synthesis of face images at different poses using only 2 input images based on a generic and deformable 3D model. Visual and infrared images are fused and then mapped on to the face model. This offers more information than either of IR or visual image alone. Based on the vivid 3D face model, images under varying poses can be synthesized and to train an face recognition system. Therefore, it can significantly reduce the size of the database and shrink registration time. The method is low cost. Experimental results shows that the proposed approach is effective and suitable for face recognition