人脸表情的实时分类

Real-Time Facial Expression Classification

提出一种基于连续Adaboost算法的人脸表情实时分类方法.使用Haar特征设计了具有连续致信度输出的查找表型弱分类器形式,构造出弱分类器空间,采用连续Adaboost算法学习出人脸表情分类器.实验结果表明:文中方法与支持向量机方法相比,对于人脸表情分类的正确率相当,而速度快近300倍,具有实时性和非常明显的应用价值.

In this paper, a real-time facial expression classification method based on real Adaboost algorithm is presented. Using Haar-like features weak classifiers of look-up-table (LUT) type, that have confidences in real values as their outputs, are designed, and correspondingly by using real Adaboost algorithm facial expression classifiers are learnt. The experimental results show that, in comparison with support vector machines (SVMs), this method achieves almost the same correction rate, and is nearly 300 times faster in speed. It could be almost in real time, and is of significance in applications.