Digit recognition method based on lip texture structure

Abstract

The invention discloses a digit recognition method based on the lip texture structure. By making use of the strong feature extraction ability of a convolution neural network and the temporal information processing ability of a long short-term memory network in deep learning, a digit is recognized based on lip movement of an object in a video through the convolution neural network and the long short-term memory network. The method is of strong robustness to the within-class difference of lip images, the head posture change, and the illumination change under a non-control environment. The problem that the lip language recognition technology is of low recognition accuracy in interactive in-vivo detection is solved. The method can be widely used in a scene equipped with a high-resolution camera, such as interactive in-vivo detection of the China's financial system.

Description

technical field [0001] The invention relates to the technical field of computer vision and pattern recognition, in particular to a digital recognition method based on lip texture structure. Background technique [0002] Using lip images to recognize numbers has been a key step in liveness detection in China's financial system. However, this task is made difficult due to the intra-class variance of lip images, variations in head poses of detected subjects, and differences in lighting in non-controlled environments. To solve these problems, extracting appropriate feature representations from video data is the key. [0003] Deep learning theory has achieved very good results in the fields of speech recognition, image target classification and detection, especially the deep convolutional neural network has a very strong self-learning ability and a high degree of nonlinear mapping. However, deep learning features based on convolutional neural networks are not capable of process...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem of low accuracy of lip language recognition technology in interactive living body detection, the present invention proposes a digital recognition method based on lip texture structure, and designs a model based on convolutional neural network and long short-term memory network

Images