I. Introduction
Biometric identification based on human physiological and behavioral characteristics has attracted much attention in identity recognition filed. Because of complex operating environment and information loss caused by artificial factors, the performance of unimodal recognition is not good, which cannot meet the demand for high-performance identity authentication. Fusion features of multimodal consist of complementary and common features of multiple modals, so multimodal fusion technologies are developing rapidly, which not only makes the extracted features more adequate but also can achieve higher recognition accuracy[1], [2], [3].