A face recognition method incorporating a plurality of improved VGG networks

Abstract

The invention provides a face recognition method integrating a plurality of improved VGG networks, Based on the existing VGG19 network, two kinds of improved VGG networks and two kinds of improved VGGnetworks with different structures are generated by deleting other convolution layers or all connection layers and their different combinations, or by changing the convolution cores and all connection layer nodes of different convolution layers. The training sample face image is preprocessed and the data set is expanded. The expanded data set is put into the improved VGG network for training, andeach network adopts different training methods to generate a variety of stable VGG models. The face images to be recognized are simultaneously put into a plurality of models for recognition, and thefinal recognition result is selected from a plurality of recognition results. The invention enables the VGG network to more effectively extract the hidden features of the deep-level face, improves thedistinguishing ability of the network to different faces, comprehensively improves the accuracy of face recognition, and adapts to the data sets of different features.

Description

technical field [0001] The invention belongs to the technical field of face recognition, and in particular relates to a face recognition method integrating multiple improved VGG networks. Background technique [0002] Face recognition technology is an identification technology based on biometrics, and it has been widely used in access control systems, monitoring systems, smart devices and other fields. Traditional face recognition methods mainly include principal component analysis, template matching, Bayesian methods, etc. The main principle is to extract different features in the face to compare the similarity of feature vectors, such as skin color features, regional geometric features, and contour features. etc. These methods are generally affected by face illumination, posture, and angle. In recent years, popular convolutional neural networks such as AlexNet, VGGNet, ResNet, etc. do not require additional manual feature vector extraction. They only need to introduce dat...

AI Technical Summary

Problems solved by technology

In recent years, popular convolutional neural networks such as AlexNet, VGGNet, ResNet, etc. do not require additional manual feature vector extraction. They only need to introduce data sets for training and learning, and automatically extract deep-level features. The disadvantage is that a large number of data sets are required. Model training requires high computer hardware computing power

[0003] So far, there have been a number of excellent face recognition network models, such as the 19-layer network VGG, the 22-layer network InceptionNet and the 152-layer network ResNet, which have achieved high recognition rates on specific data sets. The disadvantages are Complex network structure

The VGG network has only 19 layers, and the structure level is relatively shallow, but the training parameters reach more than one billion, and the training time is long. After each classification model is trained on a specific data set, it is only used for testers with certain category characteristics of this data set. The face image has good performance, and the test effect of some types in this data set or between different data sets has declined to varying degrees

Images