Living Identity Verification via Dynamic Face-Speech Recognition

In this paper, we investigate how to achieve living identity verification based on face and speech jointed recognition. Summarily, the person is required to make several expressions and read several short words, while the orders and contents are dynamically generated. Then we compare the facial components with registered ones by integrated deep neural networks (DNNs), and recognize the speeches by MVDR-MFCC features. At last, according to the results of face and speech recognition, the living identity can be verified. The innovations are as follows: firstly, the dynamic acoustic contents and faces with different expression can efficiently ensure the target is living, thus the safety of identification can be increased. Secondly, an effective component-based method is proposed for face recognition, and synthesized multiple DNNs can help reflect intensities of different components. Thirdly, MVDR spectrum is used in acoustic classification, which can effectively enhance MFCC feature. Comparative experiments demonstrate that our algorithm outperforms traditional methods in face and speech recognition accuracy, and our algorithm has particularly preponderance that it can verify the liveness of targets, so it can achieve higher security.