Bimodal 2D-3D face recognition using a two-stage fusion strategy

This paper presents a novel approach for bimodal face recognition. In our approach, faces are represented by both texture and depth images. The well known Local Binary Pattern (LBP) is used to describe the texture images. The depth faces representation is based on the Depth Local Binary Pattern, which is an extension of the the LBP descriptor allowing more discriminative power of smooth depth images. In order to perform the bimodal face recognition, a two-stage fusion scheme is proposed. It allows to take advantage of the complementarity of range and texture modalities at both descriptor (early fusion) and decision (late fusion) levels. We have conducted extensive experiments on several datasets in order to evaluate our approach. The obtained results show that our combination of texture and depth descriptors yields higher results than when taken separately or using an early/late fusion scheme.