Age Estimation Based On Convolutional Neural Network

Age estimation has important applications in the fields of criminal investigation,missing persons tracking,electronic commerce,intelligent human-computer interaction,image retrieval,etc.,so,it has received great attention from researchers at home and abroad.Although many different age estimation algorithms have been proposed,this problem is still not completely solved.When the images used to train a model and the images which the model is applied to come from different domains,for example,when the images are acquired from different regions or ethnic groups,or when the conditions of image acquisition are different,the error of age estimation is often large.To address this problem,we combine the neural network method and the traditional machine learning method.The main contribution of the thesis is as follows:?1?An age estimation method combining convolution neural network and random forest is proposed.The convolutional neural network VGG_ILSVRC₁6,which is trained on the IMDB-WIKI dataset for age estimation,is used to extract the features.For the input image,the output of each layer of the network is obtained through forward calculation,and the output of the fc6 and fc7 layers is taken as the feature maps.Then,PCA method is used to reduce the dimensions of these two feature maps.Finally,random forest models are trained for age estimation.The algorithm is validated on FG-NET and Morph-II face datasets.The experimental results show that the proposed method is superior to traditional age estimation methods.?2?An age estimation method combining convolution neural network and Support Vector Regression is proposed.The fc6 and fc7 feature maps are extracted from the convolutional neural network VGG_ILSVRC₁6,then,PCA method is used to reduce the dimensions of these two feature maps.are reduced by PCA.Finally,SVR method is used for age estimation.The algorithm is verified on the FG-NET and Morph-II datasets.The experimental results show that the proposed method is better than the traditional age estimation method,and the performance is further improved compared to the method based on convolution network and RF.?3?In addition to using the output of the FC6 and fc7 layers in the convolution neural network VGG_ILSVRC₁6 as features and using the SVR for age estimation,this paper also attempts to study the fusion of FC6 and fc7 layer output.One is featurelevel fusion,which connects fc6 features with fc7 features,and then uses SVR to estimate the age,the error slightly decreases compared with the fc6 or fc7 features alone.The other is to average the age values predicted with the fc6 feature and with the fc7 feature as the final age estimation,the error is further reduced.