摘要
微表情是一种不能自主控制和伪装的面部表情,其与诚信度的关系密切,具有持续时间短且难以识别的特征.为提高计算机自动识别微表情的准确性,提出一种基于差分能量图和中心化Gabor二值模式(centralized Gabor binary patterns,CGBP)的微表情识别方法.该方法首先利用差分法计算微表情序列的能量得到差分能量图,获得人脸面部肌肉相位的变化;其次将Gabor与中心二值模式CBP相结合,得到CGBP算子对能量图进行微表情的特征提取;最后利用ELM分类器进行微表情分类识别.在CASME微表情库上的实验结果表明,该方法比LBP-TOP、DTSA3、Gabor、VLBP、CBP-TOP算法更能有效地获得微表情序列的时空纹理特征,平均识别率为86.54%.
Micro-expression is a kind of facial expression which is autonomous and cannot be disguised. It has a close relation with credibility. Moreover,it has only a short duration and hard to be recognized. A micro-expression recognition algorithm based on the differential energy maps and centralized Gabor binary patterns (CGBP) was presented. Firstly, this algorithm uses the difference among micro-expression sequence images to calculate the energy maps and obtain the phase changes of facial muscle. Secondly, CGBP operators that combines Gabor and centralized binary patterns was proposed to extract micro-expression features. Finally, ELM classifier was used to classify micro- expressions. Experimental results on CASME micro-expression database show that compared with the state-of-the-art LBP-TOP, DTSA3, Gabor, VLBP, and CBP-TOP algorithms, this proposed method can obtain better spatial and temporal texture features and achieve higher recognition rate, which reaches 86.54% averagely.
出处
《西南交通大学学报》
EI
CSCD
北大核心
2016年第6期1222-1229,共8页
Journal of Southwest Jiaotong University
基金
国家自然科学基金资助项目(60302018)
天津市科技计划资助项目(14RCGFGX00846)
河北省自然科学基金资助项目(F2015202239)
关键词
图像处理
微表情识别
差分能量图
中心化Gabor二值化模式
ELM分类器
image processing
micro-expressions recognition
difference energy image
centralizedGabor binary patterns
extreme learning machine (ELM) elassifier