摘要
在人脸年龄特征提取方面,充分利用卷积神经网络在图像应用领域的优良特性,使用深度学习方法进行人脸年龄特征提取,采用因子分析方法进行特征降维提取鲁棒性特征。在年龄估计函数学习方面,充分利用年龄阶段性和次序性研究基于秩的年龄估计学习方法,在此基础上提出分而治之的人脸年龄估计器。利用公共年龄库FG-NET和MORPH Album 2进行实验,其结果表明,该特征提取方法比传统的年龄特征提取方法更鲁棒,分而冶之年龄估计器性能优于经典的SVM和SVR。
In the feature extraction phase,the excellent characteristics of convolution neural network（CNN）in the field of image applications were fully considered,deep learning techniques were investigated for face age description based on the CNN,and feature dimension reduction was conducted using factor analysis model.In the estimator learning phase,the stage and sequence of age were exploited for human age estimator learning based on ordinal ranking algorithm,and a human age estimator was presented based on the solution of divide and conquer.The effectiveness and accuracy of the proposed method were demonstrated with the experiments based on publicly available databases namely FG-NET and MORPH Album 2.The results demonstrate that the proposed feature is more effective than traditional hand-crafted feature,and the age estimator based on divide and conquer outperforms SVM and SVR.
作者
刘玉坤
樊爱宛
廖海斌
LIU Yu-kun FAN Ai-wan LIAO Hai-bin(College of Computer Science, Pingdingshan University, Pingdingshan 467002, China School of Computer Science and Technology, Hubei University of Science and Technology, Xianning 437100,China)
出处
《计算机工程与设计》
北大核心
2017年第11期3162-3167,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61271256)
中国博士后科学研究面上支助基金项目(2015M582355)
湖北省自然科学基金项目(ZRMS2017001419)
湖北省教育厅科学技术研究计划青年人才基金项目(Q20172805)
湖北省教育科学规划基金项目(2016GB086)
关键词
年龄估计
深度学习
特征提取
特征降维
次序秩
age estimation
deep learning
feature extract
feature dimension reduction
ordinal ranking
