摘要
为了降低样貌、姿态、眼镜以及表情定义不统一等因素对人脸表情识别的影响,提出一种人脸样貌独立判别的协作表情识别算法。首先,采用自动的人脸检测算法定位、对齐视频每帧的人脸区域,并从人脸视频序列中选择峰值表情的人脸;然后,采用峰值人脸与某个表情类内的所有人脸产生表情类内差异人脸信息,并通过计算峰值表情人脸与表情类内差异人脸的差异信息获得协作的表情表示;最后,采用基于稀疏的分类器与表情表示决定每个人脸表情的标签。采用欧美与亚洲人脸的数据库进行仿真实验,结果表明本算法获得了较好的表情识别准确率,对不同样貌、佩戴眼镜的人脸样本也具有较好的识别效果。
In order to reduce the effect of appearance, pose, glasses and non-uniform expression definition to facial expression recognition, this paper proposed a collaborative facial expression recognition algorithm based on facial appearance indepen- dently discrimination. Firstly, it selected an automatic facial landmark detection algorithm to localize and align the face region of each frame, and the peak expression faces from the video face sequence. Then, it generated an intra class variation face u- sing the peak expression face and training expression faces of an expression class, and realized the expression representation by computing the difference information between intra class variation face and peak expression face. Lastly, it adopted sparsity based classification and expression representation to decide the labels of each face expression. Simulation experimental results based on European and American faces and Asian faces show that the proposed algorithm performs better expression recognition accuracy, at the same time it shows good recognition performance to faces of different appearances and faces wearing glasses.
作者
黄超
高理平
Huang Chao Gao Liping(College of Information Science ＆ Engineering, Zaozhuang University, Zaozhuang Shandong 277160, China College of Science, China University of Petroleum, Qingdao Shandong 266580, China)
出处
《计算机应用研究》
CSCD
北大核心
2017年第9期2858-2862,共5页
Application Research of Computers
基金
山东省高校科研计划研究项目(J15LN81
J13LN56)
枣庄学院大学生研究训练计划项目(2015061)
山东省教育厅项目(J13LN56)
关键词
表情识别
稀疏表示
GABOR滤波器
峰值表情
面部标志物
expression recognition
sparsity representation
Gabor filter
peak expression
facial landmark
