摘要
针对视频人脸表情识别存在对非特定人的面部表情识别率低(约75%),有效的特征信息提取困难等问题,提出了基于分块的三维正交平面的完全局部二值模式(CLBP-TOP)的特征提取方法,实现了对任意长度视频的面部表情识别和分析。对表情视频序列进行预处理;对人脸主要区域的每个分块提取CLBP-TOP特征,得到最后的联合统计直方图特征;利用动态时间规划度量距离的最近邻规则进行表情识别;提出的表情持续时间和表情强度指数实现了对表情分析的详细描述。在CK+数据库上测试,提出的方法与LBP-TOP方法相比平均识别率提高了12. 04%,且能够对任意长度的表情视频片段进行表情识别和相关系数分析。
In view of the fact that there are problems that facial expression recognition rate of non-specific person is low( about 75 %),and extraction of effective feature information is difficult,a feature extraction method for completed local binary pattern of 3 D orthogonal planes( CLBP-TOP) based on block is proposed and facial expression recognition of arbitrary length of the video is achieved. Expression video sequence is preprocessed,and CLBP-TOP feature is extracted from each block of main face region,and the final joint statistical histogram feature is obtained. The nearest neighbor rules of the dynamic time warping is used for expression recognition. The time of duration and expression intensity scores achieve the detailed description of the expression analysis. Test on CK +database shows that the average recognition rate of this new method increases 12. 04 % compare to CLBP-TOP,and facial expression recognition and correlation coefficient analysis of any length of facial expression video clips can be performed.
作者
王小芳
王文青
魏玮
WANG Xiao-fang;WANG Wen-qing;WEI Wei(School of Computer Science and Software,Hebei University of Technology,Tianjin 300401,China)
出处
《传感器与微系统》
CSCD
2019年第4期149-153,共5页
Transducer and Microsystem Technologies
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)
天津市科技计划资助项目(15ZCZDNC00130)
河北省自然科学基金面上资助项目(F2015202239)
