由于微表情动作幅度小且持续时间短, 使其识别难度大. 针对此问题, 提出一个结合三维卷积神经网络(3D Convolutional neural network, C3D)和光流法的微表情识别方法. 所提出的方法先用光流法从微表情视频中提取出包含动态特征的光流图像系列, 然后将得到的光流图像系列与原始灰度图像序列一起输入到C3D网络, 由C3D进一步提取微表情在时域和空域上的特征. 在开放数据集CASMEⅡ上进行了模拟实验, 实验表明本文所提出的方法对微表情的识别准确率达到67.53%, 优于现有方法.
It is difficult to recognize microexpression because of its small range and short duration. To solve this problem, a micro expression recognition method based on 3D Convolutional neural network (C3D) and optical flow method is proposed. We first extract a series of optical flow images with dynamic features from the microexpression video by optical flow method, then input the obtained series of optical flow images with the original gray-scale image sequences into the C3D network, and then extract the features of micro expression in the time and space domain by C3D. Simulation experiments on the open data set CASMEⅡ show that the recognition accuracy of the proposed method is 67.53%, which is better than the existing methods.