改进深度学习块卷积神经网络的人脸表情识别  

Facial expression recognition with modified region-based convolutional neural networks

作　　者：何永强[1] 秦勤[1] 王俊鹏[2] HE Yong-qiang;QIN Qin;WANG Jun-peng(College of Computer,Henan Institute of Engineering,Zhengzhou 450007,China;School of Computer Science and Technology,Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]河南工程学院计算机学院,河南郑州450007 [2]北京理工大学计算机学院,北京100081

出　　处：《计算机工程与设计》2019年第3期850-855,共6页Computer Engineering and Design

基　　金：河南省高等学校重点科研基金项目(15A520054);河南省科技厅科技计划课题基金项目(112102310550)

摘　　要：设计一种改进的块卷积神经网络架构,并结合主动形状模型和局部二元模式映射实现人脸表情识别。采用主动形状模型定位人脸关键点,实现人脸姿态校正和感兴趣区域抽取;对校正后的图像进行局部二元模式映射,降低光照干扰;设计改进的卷积神经网络架构,对局部二元模式图像和感兴趣区域两个输入项进行学习和训练,建立分类器并实现人脸表情分类。人脸表情识别实验结果表明,该方法识别率高,运算效率较高。A modified frame of region-based convolutional neural networks was designed and used to realize facial expression re- cognition by combining active shape models with local binary pattern mapping. Active shape model was used to locate key points of face image, for correcting face posture and extracting regions of interest. Local binary pattern mapping was executed on corrected image, to reduce illumination interference. A modified frame of region-based convolutional neural networks was designed, to learn and train two input of local binary pattern image and regions of interest, and classifier was constructed and facial expression classification was realized. Experimental results of facial expression recognition show that, the proposed method has not only high recognition rate, and also higher efficiency.

关 键 词：人脸表情识别 块卷积神经网络 主动形状模型 局部二元模式 感兴趣区域池化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]