A comprehensive survey on 3D face recognition methods

Abstract

3D face recognition (3DFR) has emerged as an effective means of characterizing facial identity over the past several decades. Depending on the types of techniques used in recognition, these methods are categorized into traditional and modern. The former generally extract distinctive facial features (e.g. global, local, and hybrid features) for matching, whereas the latter rely primarily on deep learning to perform 3DFR in an end-to-end way. Many literature surveys have been carried out reviewing either traditional or modern methods alone, while only a few studies are conducted simultaneously on both of them. This survey presents a state-of-the-art for 3DFR covering both traditional and modern methods, focusing on the techniques used in face processing, feature extraction, and classification. In addition, we review some specific face recognition challenges, including pose, illumination, expression variations, self-occlusion, and spoofing attack. The commonly used 3D face datasets have been summarized as well.