| dc.description.abstract | In recent years, biometric authentication systems have been widely used in daily life, and how to improve security has always been an important topic. In the past, authentication methods based on static biometrics have become more and more easily cracked under various forgery
methods. However, methods based on sequential biometric need to be verified with sequential data, so it is relatively difficult to forge, thereby improving the security.
In this paper, an identity authentication model based on lip image and key point sequence is implemented by neural network, and the data set proposed in this paper is used for training and testing. In the general authentication experiment, the model trained in this paper obtained a
result of 8.86% HTER, which proved the effectiveness of this model for lip image and key point sequence data. In order to test whether the sequence data can achieve the purpose of improving security, we input static sequences as the fake data to the model, and obtains a result of 84.09% FAR, which shows that directly inputting sequence data is not helpful for improving security. In order to resist the static sequence attack, we calculate the frame difference of the image sequence as input. Finally, the result of 6.53% HTER is obtained in the general authentication experiment, and the result of 9.09% FAR is obtained in the static sequence attack experiment, which proves the validity and safety of the frame difference of the lip image sequence in the authentication problem. | en_US |