Face Verification Based On Convolutional Neural Network

This paper studies the application of the convolutional neural network model in face recognition in natural scenes.Traditional face recognition methods require artificially complex and time-consuming feature extraction.However,modern methods based on convolutional neural networks only need to construct an effective neural network model.Training a large number of training samples can result in higher Classification accuracy.In this approach,the network model structure plays a crucial role.Therefore,this paper mainly studies how to construct a reasonable network model structure,which enables the model to converge quickly and steadily on the training set,and finally obtain good classification accuracy.The main work of this article is as follows:(1)First,basic knowledge of gradient descending and back propagation algorithms in traditional neural networks is described.Then,the convolutional layer,pooled layer,and fully connected layer in the convolutional neural network are introduced.Some classical convolutional neural network models are introduced.(2)A new type of network model structure was designed.Using the end-to-end training mode,multiple tasks such as face alignment and face feature extraction were put into the same network model for training.This can not only use the specific field information in training signals of multiple related tasks to enhance the generalization ability,but also greatly save the training time of the model.We used the model for face recognition training in the natural environment and tested the trained model on Labeled Faces in the Wild(LFW),one of the most important face databases in the field of face recognition.Compared to DeepFace's work has significantly improved.(3)This new multi-task network model structure is trained on two loss functions of Softmax Loss and Center Loss respectively,and the advantages and disadvantages of these two models are compared.