Facial Age Estimation Based On Skin Color Classification And Deep Label

The facial image conveys rich sources of information,which can extract important information such as expression,gender,race,age,etc.As a key biological feature,age information plays an important role in pattern recognition and human-computer interaction.At present,many facial age estimation models have been successful applied to facial age estimation,but the aging of facial is also related to factors such as skin color.Therefore,this paper proposes a phased facial age estimation model based on MORPH database.The main contribution of this paper include the following aspects:(1)The current facial age estimation algorithm ignored the fact that the process of individual aging is also affected by skin color difference.This paper proposes a facial age estimation model based on skin color classification and deep label distribution learning according to different facial skin color.The model is divided into two stages.In the first stage,it classifies the faces into black skin and non-black skin,which is further feeded into the second stage's facial age estimation model to predict the age.(2)In the first stage,this paper extracts the DLGP feature,which modified local gradient mode(LGP)with a distant coefficient,and then combine it with the color feature to classify the faces into black skin and non-black skin by using XGBoost algorithm.(3)In the second stage,this paper modifies the global average pooling layer of Inception-V3 deep convolutional neural network and combines it with the label distribution learning algorithm to extract deep convolutional neural network(DCNN)feature by the fine tune technology.This stage combines deep convolutional neural network(DCNN)feature with DLGP feature,which can convey the global and local information.Finally,the regression model of the XGBoost algorithm is used for age estimation.