Research On Cross-age Face Recognition Based On Deep Learning

Nowadays,with the unceasing development of HCI technology,people more and more research on facial recognition technology,has made great achievements in every direction,can be applied to all areas of life.But face recognition has many influence factors,such as illumination,attitude change,and keep out will be a bigger impact on facial recognition rate,change is the same age,each person as the growth of the age,the change of the facial features are more or less,such as wrinkles,facial shape,and so on,this article is aimed at reducing the age changes proposed effects on face recognition.As an important attribute of the face,age is an unavoidable problem.Various features of the face will change with age,and the degree of aging of each person's facial features with age change is also different,at the same age,some people's facial features change a lot,but some people do not change,so the face aging process can not be simply processed by linear methods.In this paper,face recognition technology based on deep convolutional neural network is studied,and the following two aspects are proposed: First,this paper proposes a multi-task joint model MLFaceNet,which divides the cross-age face recognition task into two sub-tasks: age task and identity task.According to the characteristics of age characteristics and identity characteristics,the convolutional neural network is used to learn identity characteristics,and the cyclic neural network is used to learn age characteristics.Finally,the canonical loss function is introduced to balance the correlation type of the results of the two sub-tasks,and the combined results are obtained to carry out face recognition.Second,in the body member task,this paper proposes a new lightweight neural network,LightlyCNN,for learning identity features.On the basis of LightCNN neural network,it modifies its network structure and simplifies the network model,so as to obtain better identity features.The effect is better than LightCNN,and the size is also reduced.It saves computing power.Finally,the model proposed in this paper is used to conduct experiments on CACD and Morph2 data sets.Through the observation of experimental results,the recognition rate of the proposed MLFacenet model for cross-age face data is slightly higher than that of some classical algorithms.