Facial Age Estimation Based On Deep Learning

In recent years,with the rapid development of computer vision,face recognition technology has become a hot topic.The human face is one of the most important biological information of human beings,which contains a lot of important information such as gender,age,identity,race,expression and so on.Age information of human face is an important reference for identity recognition,and facial age estimation can be applied in the fields of human-computer interaction,video surveillance and image video intelligent analysis.Therefore,the research on age estimation has attract more researchers' attention.Over the past decades,researchers have been making unremitting explorations and put forward different algorithms,however these algorithms are still limited by conditions at that time,so age estimationhave much room for improvement.In recent years,deep learning has made tremendous progress in all fields.Especially,field of image recognition has also made breakthrough research progress on deep learning.And the task of identifying the age of face images has also gradually achieved better results than the traditional algorithms.Based on the theory of deep learningwe propose a facial age estimation method based on Deep Fusion Network(DFN).The main contents of this paper are as follows:(1)We study the commonly used age estimation and introduce the research status of the two parts of age estimation.In the part of feature extraction,we summarize the traditional feature extraction method and the feature extraction method based on deep learningin detail.In the part of age estimation,we introducethe classification and regression.Then we have introduced the evaluation index of age estimation systemand compared the advantages and disadvantages of all age estimation algorithms.(2)We thoroughly understand the development of deep neural networkand comb the development process of deep learning.Then,we list some achievements of research and practical applications in deep learning.In addition,weintroduce the network structure principle and training methods of DBN and CNN.And we make a series of age estimation experiments for the two representativenetworks,with the input samples include original images and the processed images,then we analysis the experiment results according to the two networks;(3)We propose a facial age estimation method based on Deep Fusion Network(DFN),which adopts the idea of stacking multiple CNNs and a DBN to extract and fuse facial features for age estimation.Multiple CNNs are used to extract deep non-linear features from multiple local regions of face images.And these features are fused by a DBN model to generate the final value of estimated age.(3)We use a network-by-network hierarchical training principle to train the DFN.The multiple CNNs are initially trained on the larger scale face database.For reducing overfitting,the pre-trained CNNs are gradually fine-tuned on the limited face age database.After unsupervised pre-training of the DBN with the concatenated CNN features,the DFN is globally fine-tuned based on a regression loss function.We conduct extensive experiments to evaluate our proposed method.The experiments are performed on two popular used benchmarks: FG-NET database and MORPH II database.Experimental results on two databases show that the proposed DFN-based method is an effective deep architecture for facial age estimation and achieves a competitive(MAE = 3.42 and MAE = 4.14)performance in comparison with state-of-the-art methods.The proposed DFN is an effective deep learning model for age estimation and can achieve a competitive result compared with state-of-the-art methods.