深度卷积神经网络对人脸年龄的分类  

Deep Convolutional Neural Network for Age Classification from Facial Images

作　　者：陈莉明[1] CHEN Liming(School of Physics and Electronic Engineering,Leshan Normal University,Leshan Sichuan 614000,China)

机构地区：[1]乐山师范学院物理与电子工程学院,四川乐山614000

出　　处：《乐山师范学院学报》2018年第12期1-4,44,共5页Journal of Leshan Normal University

基　　金：国家自然科学基金资助项目"基于深度卷积神经网络的盲图像质量评价"(61701351);四川省教育厅科研项目"deep learning在图像处理中应用的研究"(16ZB0295);乐山市重点科技计划项目"deep learning在人脸识别中的应用研究"(16GZD032)

摘　　要：为了更好地进行人脸年龄分类,文章提出了一种基于深度卷积神经网络的年龄分类方法。该方法从VGG-Face倒数第二激活层输出年龄特征,然后采用概率协同表示分类器(ProCRC)进行年龄分类。通过在两个数据集上的验证和比较可知由深度卷积神经网络和ProCRC组成的混合深度卷积神经网络能提高年龄分类的性能;通过将倒数第二激活层和倒数第一激活层所输出的特征在同一分类器上进行分类的结果进行比较得出对倒数第二激活层输出特征进行分类可提高人脸年龄分类的精度。In order to make a better age classification from facial images,an age classification method based on deep convolutional neural network is proposed.By this method,the age features are output from the penultimate activation layer of VGG-Face and ages are classified by a probabilistic collaborative representation based classifier(ProCRC).Through validation and comparison on two data sets,it can be seen that the hybrid deep convolutional neural network composed of deep convolution neural network and ProCRC can improve the performance of age classification.Meanwhile,it also can be seen that the comparison of the age classification accuracy of the output of the penultimate activation layer and that of the output of the last activation layer based on the same classifier shows that the former is higher.

关 键 词：年龄分类 人脸年龄特征 深度卷积神经网络 概率协同表示分类器 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术]