RGB-D动态序列的人脸自然表情识别  

Spontaneous Facial Expression Recognition Based on RGB-D Dynamic Sequences

作　　者：邵洁[1] 董楠[2] 

机构地区：[1]上海电力学院电子与信息工程学院,上海200090 [2]中国科学院上海高等研究院城市公共安全研究中心,上海201203

出　　处：《计算机辅助设计与图形学学报》2015年第5期847-854,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61302151);上海市自然科学基金(13ZR1455100)

摘　　要：区别于以二维静态图像为对象的传统人脸表情识别,提出一种针对RGB-D动态图像序列分析的人脸自然表情自动识别算法.首先针对预处理后的RGB-D表情图像序列提取四维时空纹理特征作为局部动态特征;再利用慢特征分析自动检测表情序列的峰值图像,并提取脸部三维几何模型为全局静态特征;最后结合动、静态特征,经主成分分析降维后输入条件随机场模型完成特征训练和表情识别.经由BU-4DFE人脸表情库验证表明,该算法不但比传统静态表情识别算法和其他动态算法具有优越性,而且能够针对自然展现的表情实现自动识别,为今后算法的实用化提供了可能.Different from traditional facial expression recognition methods based on 2D static images, a spontaneous facial expression recognition algorithm is proposed for RGB-D image sequences. After pre-processing on image alignments and normalization, 4D spatio-temporal texture data are extracted as dynamic features. Then Slow Feature Analysis method is applied to detect the apex of the expression, so that 3D fa-cial geometrical model of the apex image is built and used as the static feature. With the combination of these two kinds of features and the dimensional reduction by PCA, Conditional Random Fields is applied to train and classify the features in the end. A lot of experiments were performed based on BU-4DFE facial ex-pression database. It has been verified that our algorithm not only outperforms traditional static facial ex-pression recognition methods and many other dynamic facial expression recognition methods, but also could recognize spontaneous expression automatically, which makes it possible for further practical applications.

关 键 词：表情识别 动态表情序列 四维纹理特征 三维几何模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]