基于多核学习的表情识别方法

Facial Expression Recognition Based on Multiple Kernel Learning

传统的表情识别方法大多采用一种特征提取方法如Gabor特征、LBP特征等进行表情识别,其中每种特征提取方法对人脸表情特征的描述各有优缺点,单纯利用一种表情特征进行表情识别,识别率往往不高。提出一种基于多核学习的多种特征有效结合的表情识别方法,以兼顾不同特征对表情识别的作用。利用日本表情数据库JAFFE进行方法的仿真实验,结果表明:基于多核学习的表情识别方法识别率高于传统的基于单核方法。

In traditional facial expression recognition, only one kind of features extraction method such as Gabor feature, LBP feature, etc. Each kind of features show both advantage and disadvantage in facial expression description, and it is hard to reach high recognition rate if only one type of features is used. In this paper, multiple kernel learning （MKL） based facial recognition method is proposed, which combines multiple features effectively, and considers functions of different expression features fully. Experimental results on the JAFFE database show that the proposed method performs better than the traditional single kernel ones.