基于三维卷积神经网络与支持向量机的微表情自动识别

Facial Micro-Expression Auto-Recognition Using 3D Convolutional Networks and Support Vector Machine

不同于常规的人脸表情,微表情是一种在人们试图压抑或隐藏真实情感时产生的非常短暂的、不能自主控制的人脸表情序列。微表情识别可以应用于审犯测谎、商务谈判等场合。微表情持续时间短,动作幅度小等特点,使得人们难以用肉眼察觉和感知,目前微表情识别准确率较低。为了解决这个问题,提出采用三维卷积神经网络与支持向量机(C3DSVM)的方法来对微表情进行自动识别,首先通过C3D提取微表情在时域和空域上的特征,然后使用SVM分类器进行分类。在CASME2数据集上做实验,实验表明该方法的识别准确率比其他前沿方法高,达到88.79%。

Unlike normal face expression, micro-expression is a very brief, involuntary facial expression that produces when people try to suppress or hide real emotions. It can be used in trials such as polygraphs and business negotiations. The short duration of micro-expressions and the small intensity of motion make it difficult for people to aware and perceive with the naked eye. Currently, the micro-expression recognition accuracy is low. In order to solve this problem, proposes a three-Dimensional Convolutional Neural Network and Support Vector Machine method (C3DSVM) to recognize micro-expression. Firstly spatio-temporal features from micro-expressions are extracted, and then are classified by SVM classifier. Validates the proposed method on CASME2 dataset, and achieves 88.79% on accuracy, which is higher than other state-of-the-art methods.