Research On Face Age Estimation Algorithm Based On Convolutional Neural Network

With the continuous development of computer vision,face detection and recognition technology has gradually become mature and has been widely used.As an important attribute of face,age estimation attracts more and more attention.It is a research focus in the fields of pattern recognition and machine learning.It has great potential value and can be widely used in the fields of security control and human-computer interaction.In this paper,we mainly study the problem of face age classification,that is,predicting the age range of face images.Most of the current methods use traditional handcraft features to describe age information,and these low-level features are often inadequate.In order to obtain more effective age characteristics,we use convolutional neural network as a classification model.Firstly,the age classification model based on convolutional neural network is proposed.In recent years,deep learning has been developed rapidly.As one of the important model,convolutional neural network can take the image pixel as input and extract features from local information blocks.These features make convolutional neural network popular in computer vision,especially in image processing.Therefore,an eight-layer convolutional neural network is constructed for the age classification problem.And we propose an improved loss function for the ordering of the age labels,an additional distance term is added to the original cross-entropy function so that when the distance between the predicted class and the true class is larger,the loss value is larger.In addition,some useful methods such as Dropout,Batch Normalization and ReLU activation function are used to train the network.Finally,we evaluate our method on the Group dataset for age estimation,the experiment results demonstrate the performance of our method against the traditional methods based on handcrafted feature extraction.Secondly,the age classification model based on ensemble convolutional neural network is proposed.As a useful weapon in the field of machine learning,ensemble learning can effectively improve the generalization ability of the model.Aiming at the insufficiency of the generalization ability of the above convolutional neural network model,the age classification model based on the ensemble convolutional neural network is further proposed.Three convolutional neural networks with the same structure and different inputs are weighted integrated,The result of the ensemble is the final classification result.The inputs of the three networks are the grayscale image,the histogram equalized image,the texture feature extracted by Local Binary Pattern respectively.The three neural networks play a complementary role,using the original grayscale image retains the information of the face;the image processed by the histogram equalization enhances the contrast,and the new image is more clear than the original one;Finally,considering the importance of wrinkle information of age classification,the classification result of LBP texture feature is added.Experiment results show that this ensemble learning model effectively improves the accuracy and the generalization ability of the age classification.