Research On Face Recognition Algorithm Based On Adversarial Learning

With the improvement of living standard,people pay more and more attention to their appearance,many people will use makeup to improve their appearance.However,most of the existing face recognition and face verification algorithms are based on facial feature information.Due to makeup,facial feature information is changed,which will eventually lead to the feature extraction network fails to extract effective features,which will greatly reduce the accuracy.The existing methods of face recognition and verification generally make the model more robust by increasing the number of samples in the training set or the number of network layers.However,increasing the number of training set will lead to longer training time and lower training efficiency.However,increasing the number of network layers will make the model parameters larger and the training speed slower.In addition,relevant studies have shown that the model will degrade after the number of network layers continues to increase,and the effect will decline instead of rise.On this foundation,this paper introduces adversarial learning into the face image space that needs to be recognized and the intermediate feature space of the face recognition network,and proposes two methods of face verification with makeup based on adversarial learning.The specific contents are as follows.(1)A method of face verification with makeup based on makeup directed generation was proposed.First,a neural network was trained to evaluate the makeup features of each face image.Secondly,the network is used to constrain the generation process of a residual generation adversarial network,so that the network can achieve the targeted generation of makeup.A face without makeup is converted into a face with characteristic makeup to eliminate the information difference of face image caused by makeup.After that,it is used to train a residual network to achieve the goal of face verification.The generation experiment on MYP data set proves that the method proposed in this paper can achieve the targeted generation of makeup,and the generated image is better.Also in the face verification experiment on this data set,the correct acceptance rate of this method is 0.3628 if the error detection rate is less than 0.01.In the experiment of simulating the actual situation,the accuracy rate reached 84.24%,which was the highest among all the existing methods,indicating that this method can improve the existing face verification model and improve the robustness of makeup.(2)A method of face verification with makeup based on middle layer adversarial learning was proposed.Two residual networks are trained to extract face features with and without makeup.In the feature space extracted from the two networks,the triplet loss is used to constrain the network so that they can extract the corresponding face features.In addition,on the basis of the feature space extracted from the middle layer of the network,a discriminator is introduced to make the two residual networks extract features irrelevant to makeup by means of adversarial learning.In experiments on MYP data sets,the correct acceptance rate of this method is 0.4266 with a false detection rate of less than 0.01.At the same time,in the experiment on the MIFP data set,the correct acceptance rate of the method is 0.2563 if the error detection rate is not more than 0.001.In the experiment which simulates the actual situation,the accuracy reached 75.64%.The two indexes are the highest in the existing methods.It shows that compared with other methods,this method is not easy to attack.