Age Estimation Based On Face Image

In recent years,automatic information processing technology related to face images has developed into an important research topic in the field of information technology.Face age estimation is a technique of automatically estimating age based on face images,which belongs to the prediction task of combining biological characteristics with computer vision.It is divided into physiological age estimation and apparent age estimation.Face age estimation is facing some difficulties and challenges.On the one hand,face aging is an uncontrollable natural process,and different people have different aging patterns.On the other hand,age estimation lacks a complete and sufficient data set.From the current research situation,there is room for further improvement in both physiological age estimation and apparent age estimation.This paper focuses on the optimization of algorithm and network,and puts forward two models for predicting face age,as follows:(1)Physiological age estimation based on improved Label Distribution LearningThe facial age characteristics of the same individual with similar age are similar,and the age estimation based on Label Distribution Learning is a method designed by using this characteristic.It realizes the change of learning task from single target prediction to age mark distribution vector prediction,which solves the problem of incomplete data in face age estimation to a certain extent.However,the existing Label Distribution Learning based on the maximum entropy regression model can not build a unified prediction model of label distribution and long calculation time.Another algorithm based on neural network is prone to over-fitting and the structure of neural network is not easy to understand.To solve these problems,a new method based on kernel partial least square regression model is proposed to solve the problem of face age estimation.The kernel partial least square regression model has no precondition for data distribution and can solve nonlinear problems.The experimental results of FG-NET and MORPH II datasets show that compared with other methods,the method has smaller age estimation error and improves the calculation efficiency.(2)Apparent age estimation based on improved residual networkAge estimation method based on deep learning is widely used,but with the increase of network depth,there will be gradient instability,network degradation and increasing number of parameters.Aiming at the problems of the existing convolutional neural network model algorithms,such as the lack of expression ability of face age features,low recognition accuracy and large amount of model parameters,a residual network face age estimation method integrating attention is proposed.Firstly,the original convolutional block attention module(CBAM)is optimized and improved,and the channel importance weight is obtained by feature fusion and retraining.The mutual information of the two parts is pooled and compressed by network layer hybrid calculation,and the representation of important channel features is enhanced,so as to improve the fitting ability of the model.Then,the residual structure of the improved CBAM and RESNET model is fused to construct a new residual module;This module can avoid the influence of CBAM on the back propagation of shortcut connection.Finally,Softmax is used to classify and obtain the final estimated age.Experiments on APPA-REAL and LAP2015 apparent age data sets show that this method achieves better average absolute error and cumulative error index than the original Res Net model method.Compared with other related methods,it also proves its effectiveness.