基于光流与LBP-TOP 特征结合的微表情识别

微表情区别于普通的面部表情, 具有持续时间短、面部强度低的特点, 往往难以有效识别, 制约了该领域的研究。针对上述难点, 提出一种新颖的特征结合方法。采用全局光流技术在相邻帧间进行计算, 得到微弱光流, 通过传递前后各帧的运动信息, 在相隔多帧的两幅图像间体现更为明显的变化, 解决了短历时和动作微弱的难题; 将光流特征与LBP-TOP(Local Binary Patterns from Three Orthogonal Planes)算子提取的时空局部纹理特征相结合, 补充描述人脸大多数区域的细节信息。选择随机森林分类器进行实验, 实验结果表明, 两种特征具有很好的互补性, 在CASMEII 数据库下, 能识别5 类情感, 准确率由40. 50%提高至64. 46%, 类间区分度也有相应改善。

Micro-Expression Recognition Based on Feature Combination of Optical Flow and LBP-TOP

Different from ordinary facial expression, micro-expression has the characteristics of short duration andlow intensity, which is often difficult to be identified effectively, resulting in the limitation of research in this field. Dealing with these difficulties, a novel feature combination method was proposed. Calculation using the global optical flow technology was carried out in the adjacent frames to obtain faint optical flow. The motion information of each two adjacent frames was passed by, so that the changes were significant between the images of two frames at a distance, solving the problem of short duration and weak action. The optical flow feature and space-temporal local texture feature extracted by LBP-TOP(Local Binary Patterns from Three Orthogonal Planes) operator were combined to make supplement for describing most region details of the faces. The random forest classifier was selected, and experimental results show that the two features have a very good complementary, in the CASMEII database. The method is able to identify five types of emotion, the accuracy is promoted from 40. 50% to 64. 46%, and the category discrimination is improved accordingly.