Multimodal Biometric Authentication System for Smartphone Based on Face and Voice Using Matching Level Fusion
Xinman Zhang, Dongxu Cheng, Yixuan Dai, Xuebin Xu · 2018
Recently, the multimodal biometrics authentication method has been widely studied since it can overcome the deficiency of unimodal biometric. In this paper, a multimodal biometric authentication system for smartphone based on face and voice is developed to solve the problem where the single biometric is easy to be stolen. In view of the low CPU performance of the smartphone, which can't store and quickly process large amount of data, face detection is carried out after image acquisition and redundant background image is discarded to reduce futility information effectively. In addition, the traditional LBP operator is improved to enhance the robustness of the authentication system and the traditional endpoint detection technology is optimized to deal with the mute and transition information in the voice stream. For the authentication process, an adaptive fusion strategy based on the matching level is proposed. The simulation experiments based on PC show that the authentication accuracy of our algorithm on the benchmark database reaches 100% when the training sample number is 5, and the single authentication time is about 341ms. A lot of test experiments on smartphone show that our system achieves perfect authentication effect and meets the requirements in practical application.