基于C均值K近邻算法的面部表情识别

Facial expression recognition based on C-means and K-nearest neighbor algorithms

随着人工智能与模式识别技术的不断发展,面部表情识别在智能人机交互中发挥着越来越重要的作用.通过对人的面部表情分类的研究,提出了一种使用C均值聚类、K近邻算法的面部表情分类方法.对参加训练的表情图像先进行Gabor小波变换,然后使用Fisherface判别分析方法进行变换,求得特征空间.再将已进行Gabor变换的标准表情图像投影到特征空间,进行C均值聚类得到子类表情模板.对于一幅待识别的表情图像,使用K近邻算法与子类表情模板比较,将该表情图像分类.使用该方法,在公开的日本女人表情人脸库上实测达到了95·8%的识别率.

With the rapid development of artificial intelligence and pattern recognition, facial expression recognition plays an important role in intelligent human-machine interaction. In this paper, a facial expression classification method is presented which uses a C-means and K-nearest neighbor algorithm as the basis of analysis for the classification of facial expressions. First the images to be analyzed are transformed with Gabor wavelets, and then Fisherface discriminate analysis is performed to generate a feature space. Next, the images which were transformed with Gabor wavelets are projected into the feature space and C-means clustering performed on the projected images to generate sub-expression templates. Finally, the type of expression is identified by comparing the input expression images with the sub-expression templates by using a K-nearest neighbor algorithm. Experiments on the public Japanese female facial expression database show that the method proposed in this paper can achieve a 95.8% recognition rate.