Multimodal Biometrics Based on Convolutional Neural Network by Two-Layer Fusion

Abstract

With the growing demand for information security, biometric recognition technology has been widely used in daily life. Due to the disadvantages of unimodal biometric system, multimodal biometric recognition has become more popular. In view of the excellent performance of deep learning theory and method in various recognition tasks, this paper proposes a multimodal biometric recognition method based on three biometric modalities of face, iris and palmprint under the framework of deep learning. Firstly, in order to investigate the effect of model structure on recognition accuracy, we design the different structures of Convolution Neural Network (CNN) for unimodal biometric recognition. Then, combined with the results of unimodal recognition, we construct a CNN model based on two-layer fusion for multimodal biometric recognition. Specifically, a variety of feature fusion strategies are introduced for exploring the influence of fusion methods and mechanism on recognition performance. Finally, a large number of experiments are carried out on three standard databases. The experimental results show that the proposed multimodal recognition method has higher recognition accuracy than unimodal recognition. Moreover, the two-layer fusion mechanism can further improve the multimodal recognition performance.