一种基于LNMF像素模式纹理特征的表情识别

Facial expression recognition using LNMF based on PPBTF

采用一种基于像素模式纹理特征(PPBTF)的人脸特征表示方法对人脸图像进行了特征提取.首先,将原始的灰度图像转化成能够表征纹理信息的模式图,并且通过在特征窗内统计每一模式的像素个数得到其中心像素的特征矢量,然后将由局部非负矩阵分解(LNMF)得到的基本方程作为模板进行模式匹配.同时,将Adaboost和SVM结合起来,用做表情识别的分类器.最后,通过基于Cohn-Kanade数据库的实验证明了以LNMF基函数作为模板的PPBTF对表情识别具有较高的判别能力,并由基于PIE图像库等其他图像库的实验进一步验证了PPBTF对光照不敏感的特性,充分说明所提出的人脸表征方法的有效性和鲁棒性.

A new sort of face representation method -- pixel-pattern-based texture feature （PPBTF） is adopted to extract features from facial images. A gray scale image is firstly transformed into a pattern map where edges and lines can be used for characterizing the facial texture information. Secondly, based on the pattern window, a feature vector is comprised of the numbers of the pixels belonging to each pattern. Then, the image basis functions obtained by local non-negative matrix factorization （LNMF） are used as the templates for pattern matching. Additionally, Adaboost and support vector machine （SVM） are adopted to classify facial expression. Experiments on the Cohn-Kanade database illustrate that the PPBTF combined LNMF is efficient for facial expression recognition, and experiments on the PIE and other database also verify the PPBTF＇s insensitivity to the lighting condition. Conclusively, the face representation method proposed is proved to be efficient and robust.