摘要
将Zernike矩和小波矩运用于面部表情识别问题 ,分别计算了面部图像的Hu矩、Zernike矩、Haar矩、Shannon矩和B样条矩 ,以模式识别中常用的类间距作为依据 ,提取了面部图像的各种矩的最好特征和次好特征 ,并对Zernike矩和B样条矩的识别能力和抗噪性进行了比较 .实验证明 :用Zernike矩作为面部表情特征 ,其识别率在特征数取 5个时能达到 95 % ,B样条矩在特征数取 2个以上时识别率能达到 1 0 0 %
Wavelet moment invariants are applied to the facial expression recognition. Zernike moment invariants, Haar moment invariants, Shannon moment invariants and B spline moment invariants were calculated respectively in human facial expression images. According to the criteria of class distance, the best and the better feature of several moment invariants were extracted from facial expression images. And the recognition ratio of Zernike moment invariants were compared with that of B spline moment invariants. Experiment has proved that the recognition ratio of Zernike moment invariants can be obtained 95% when 5 features are used, and the recognition ratio of B spline moment invariants can be obtained 100% when more than 2 features are used.
出处
《东南大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2004年第4期557-560,共4页
Journal of Southeast University:Natural Science Edition
关键词
表情识别
小波矩
ZERNIKE矩
B样条矩
facial expression recognition
wavelet moment
Zernike moment
B spline moment