基于四元数局部编码和卷积网络的表情识别  

Expression recognition based on quaternion local coding and convolutional network

作　　者：薛志毅 邵珠宏[1,2] 江筱 赵晓旭 尚媛园[1,2,3] XUE Zhi-yi;SHAO Zhu-hong;JIANG Xiao;ZHAO Xiao-xu;SHANG Yuan-yuan(College of Information Engineering,Capital Normal University,Beijing 100048,China;Beijing Advanced Innovation Center for Imaging Theory and Technology,Beijing 100048,China;Beijing Engineering Research Center of Highly Reliable Embedded System,Beijing 100048,China)

机构地区：[1]首都师范大学信息工程学院,北京100048 [2]北京成像理论与技术高精尖创新中心,北京100048 [3]高可靠嵌入式系统北京市工程研究中心,北京100048

出　　处：《计算机工程与设计》2020年第2期507-512,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目（61601311）;北京市属高校高水平教师队伍建设支持计划基金项目（CIT&TCD20170322）;北京市优秀人才基金项目（2016000020124G088）;北京市教委科研计划基金项目（SQKM201810028018、KM201910028018）

摘　　要：为充分利用彩色图像的颜色信息和通道之间的关联性,提出一种联合四元数矩阵相位信息和幅值信息的特征提取方法,结合卷积神经网络(CNN)进行表情识别。将彩色表情图像表示为纯四元数矩阵并进行Clifford平移,对相位和幅值分别进行局部二值模式(LBP)编码,提取多尺度融合的图像特征,将特征输入CNN进行训练并分类。实验结果表明,该算法在RafD和MMI表情库上的识别率分别为79.42%和93.28%,相比其它表情识别算法,识别率更高且识别效果稳定。To make full use of the color information of color image and the correlation among different channels,an approach of feature extraction combining the amplitude and phase information of quaternion matrix was proposed,which was used for expression recognition together with convolutional neural network(CNN).Color facial images were represented into pure quaternion matrices and operated with Clifford translation.The weighted phase and amplitude information were respectively encoded using LBP to extract the multiscale fused features.The fusion features were served as the input of CNN for training and classification.Results of experiments performed on RafD and MMI facial expression databases show that the recognition rates are 79.42%and 93.28%respectively.Compared with other facial expression recognition algorithms,the proposed algorithm has higher recognition rate and its recognition performance is stable.

关 键 词：人脸表情识别 四元数表示 特征提取 局部二值模式 卷积神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]