结合局部二值模式和卷积神经网络的人脸美丽预测  

Facial Beauty Prediction Combined Local Binary Pattern and Convolution Neural Network

作　　者：甘俊英[1] 谭海英 项俐 翟懿奎[1] 曾军英[1] GAN Jun-ying;TAN Hai-ying;XIANG Li;ZHAI Yi-kui;ZENG Jun-ying(School of Information Engineering, Wuyi University, Jiangmen, Guangdong 529020, China)

机构地区：[1]五邑大学信息工程学院,广东江门529020

出　　处：《信号处理》2019年第1期150-158,共9页Journal of Signal Processing

基　　金：国家自然科学基金项目(61771347;61072127;61372193);广东省特色创新类资助项目(2015KTSCX145);广东省自然科学基金项目(S2013010013311)

摘　　要：卷积神经网络(Convolution Neural Network,CNN)用于人脸美丽预测,能学习到深层次的特征表达,但提取的是全局特征,忽略了人脸的局部信息,因此,泛化能力不强。为此,本文提出一种结合局部二值模式(Local binary pattern,LBP)和卷积神经网络的人脸美丽预测算法。首先,利用数据增强技术扩大数据库规模;其次,将LBP纹理图像和原始灰度图像进行通道融合;再采用1×1卷积操作进行通道特征图的线性组合,从而实现网络跨通道的信息整合,提升人脸美丽预测精度。基于大规模亚洲女性人脸美丽数据库(Large Scale Asian Fe-male Beauty Database,LSAFBD)的实验结果表明,该算法在分类和回归预测中均取得了较好效果,优于其他模型的人脸美丽预测算法;表明在卷积神经网络中加入纹理图像能有效提升人脸美丽预测精度。Convolution Neural Network (CNN) for facial beauty prediction, can learn the deep feature expression, but can extract the global feature and neglect the local information of the face. Therefore, it has poor generalization ability. In this paper, a facial beauty prediction algorithm combined Local binary pattern (LBP) and CNN is presented. Fistly, data augmentation technology is utilized to expand the scale of the database. Secondly, the LBP texture image is channel-fused with the original grayscale image, and then the linear combination of channel feature maps is implemented by a 1×1 convolution operation. Thus the cross-channel information fusion of the network is realized, so as to improve the accuracy of facial beauty prediction. Experimental results based on the Large Scale Asian Female Beauty Database (LSAFBD) show that the algorithm presented has good prediction ability in the classification and regression, which is superior to the other models for facial beauty prediction, and demonstrate that adding texture images to CNN can effectively improve the accuracy of facial beauty prediction.

关 键 词：人脸美丽预测 局部二值模式 卷积神经网络 LSAFBD人脸美丽数据库 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]