Research On The Algorithm Of Face And ID Photo Verification Based On Deep Learning

A face and ID card photo verification is to verify identity between the ID card photo and the face image of the ID card holder.With the diversity of people's social activities,fast and accurate identification of identity information is becoming more and more important.Because ID card images have special features in face pose and shooting conditions,etc.The face recognition algorithm is applied to the comparison between ID card photos and face images,but this often leads to low recognition rate.This paper is based on deep learning theory,The goal is to improve the accuracy of face recognition under the condition of variable facial posture and age span.The algorithm of face and ID photo verification based on deep learning is proposed.The main work is as follows:(1)A face recognition algorithm based on multi-pose face recognition by dynamic loss weights is proposed.The weight of loss function is dynamically adjusted according to different face posture tags.In order to improve the accuracy of multi-pose face recognition,different sub-models of face recognition are trained according to different facial pose,and each sub-model is finally fused into multi-pose face recognition model.(2)The feature of eyes,nose and mouth is the key of human identification.A face and ID card photo recognition method based on LBP and convolution neural network was proposed.LBP features of eyes,nose and mouth are fused with face features extracted from deep convolutional neural network.Enhance the feature representation intensity of key parts of the face,reduce the difference of features caused by time span,and improve the recognition accuracy.(3)In order to improve the robustness of face features extracted from face recognition model.A multi-pose face and ID card photo identification algorithm based on double input double loss convolutional neural network is proposed.The ID card dataset and CASIA are alternately input into the convolutional neural networkand supervised by different loss functions.At the same time,combining the idea of multi-pose face recognition algorithm based on dynamic loss function,the multi-pose face recognition model is trained,and improve the accuracy of face and ID card photo verification.The experimental results on a variety of open face recognition test sets and self-collected face and ID card photo test sets show that the proposed algorithm can improve the accuracy of multi-pose face and ID card photo verification under unconstrained conditions.