Research Of Face Age Classification Base On Ensemble Convolutional Neural Network

With the development of Internet Technology,the Internet has played a very important role in our daily life.The rapid development of human-computer interaction technology,intelligent human-computer interaction has become has become a focus recently.The areas of images recognition such as human face recognition,human face age recognition,gender recognition andemotion recognition,has got more and more attention.Particularly,the face age classificationhas significant potential value in public security investigation,image retrieval,intelligent monitoringand so on.This paper mainly studies face age classificationunder unfiltered faces.Unlike traditional face classificationunder standard images which is collected in laboratory environment,our age classification under unfilteredfaces,has more realistic meaning and application value.In this paper,convolutional neural network,which is a method of deeplearning,is used to solve the problem of face age classfication.However,training convolutional neural network requires a lot of training data,and there is a lack of database of face age classification under unfiltered faces.Thus,this paper uses image rotation method,fine-tuning RGB channelsand add Gaussian noise in face images to augment the training set,but testing set remains unchanged.Based on the traditional convolution neural network,this paper constructwd a neural network model for face recognition called AGE_CNN,this model gain a highest accuracy rate than the existing paper increased by 1.1% and 1.6% on Adience and Gallagher datasetseparately.Experiments show that using rotation method,fine-tuning RGB and adding channel Gaussian noise disturbances in face images to augment the training set,the accuracy in Adience and Gallagher dataset was highly improved by 1.4% and 1.2 % separately than not.This paper also research on the feasibility face age classification using ensemble convolution neural network.According to the characteristics of the convolution neural network,this paper constructed a number of convolution neural networks withdifferent number ofkernelsor different kernel's size.By a large number of experiments on these convolution neural networks,this paperselected six better integrated neural network model convolution included B_CNN₁,B_CNN₂,B_CNN₃,B_CNN₄,B_CNN₅,B_CNN₆ as based classifiers of emsemble classifier.Experiments show that,the accuracy on Gallagher and Adience dataset improve 3.4% and 4.6% than the existing paper.Finally,base on the experimental results above,this paper designed and developed a face age classification system which was performed well in our test.