Research On Facial Age Estimation And Aging Synthesis

In recent years,facial analysis has become one of the most popular topics in computer vision with the opening of human-computer interaction,security monitoring and entertainment market.Age is a significant reference for personal identification.Although face age estimation and aging synthesis technology have been greatly developed in recent years,there are still many difficulties and challenges to be dealt with.On the one hand,face aging is a slow and irreversible changing process of facial shape and texture,following some common patterns of the human aging process as well as affected by gender,race,lifestyle and other factors.On the other hand,the existing age labeled datasets fails to meet the demands for age estimation and aging synthesis.Therefore,continuous efforts and innovations should be made for age estimation and aging synthesis.In this paper,the author mainly uses Convolutional Neural Network(CNN)and Deep Convolutional Generative Adversarial Networks(DCGAN),to study age estimation and aging synthesis of face images respectively.The main research work in this paper is as follows:(1)A multi-label age estimation based on CNN is proposed.Different from most traditional age estimation methods,this method has no explicit feature extraction and age estimation stage,realizing end-to-end age estimation by training a multiple output convolution neural network.In the age estimation algorithm,in-depth analysis on the characteristics of age estimation is made,different from the general classification.This method changes the single label to multi age label by taking fuzzy information in consideration,in which,a sample label can contribute to the learning of its adjacent age label.The output layer of the convolution neural network where a binary classifier is designed for each age label,then the age estimation problem is transformed into a series of binary classification sub-problems.Finally,using the ordinal information of face age to estimate age in local age range,which is calculated based on the binary classifier output.Extensive experimental results on face aging datasets show that our method achieve superior performance.(2)An ageing synthesis method based on DCGAN is proposed on the basis of the powerful image generative ability of generative adversarial networks(GAN)which mapped the face feature vector and age conditions to the face image through a generator.In the network structure,a convolution encoder is designed to extract the personality feature vector of the face,and combines the age estimation network that provides age estimation for the input image and the synthetic image,forcing the network to synthesize the aging image of the specified age group.To generate clearer faces,three more loss items,namely image feature loss,image pixel space loss and age loss,are added to objective function besides GAN adversarial loss.The final objective function is the weighted sum of the loss terms.At last,analysis and comparison of experimental results were given to illustrate the efficiency and feasibility of this algorithm.