摘要
针对新生儿的疼痛表情识别,提出了将Gabor小波变换与改进的KDA相结合的特征提取方法。首先对新生儿面部图像进行Gabor变换,然后针对变换后的Gabor特征,用一种改进的核鉴别分析方法对它进行二次特征提取。该方法从根本上解决了表情识别中因小样本问题而引起的核类内离散度矩阵(kernel within-class scatter matrix)奇异性的问题。最后,对提取的特征用支持向量机进行了疼痛表情的分类识别。实验结果表明,此表情特征提取方法能够显著改善表情识别系统的性能。
In this paper,we propose a facial expression feature extraction method which combines Gabor wavelet transformation with improved KDA in order to recognize the neonatal facial expression of pain. At first, the neonatal facial image is transformed by Gabor wavelet. Then, the Gabor features are extracted with an improved kernel discriminant analysis method in order to solve the singularity problem of kernel within-class scatter matrix which is caused by the small sample size problem of expression recognition. Based on Support Vector Machine classifier, experimental results show that：the proposed methods of facial expression feature extraction can greatly improve the performance of recognition system.
出处
《南京邮电大学学报（自然科学版）》
EI
2008年第5期1-6,共6页
Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition
基金
Asian-Swedish Research Links Programme(348-2005-6434)
南京市留学回国人员科技活动择优资助经费(TJ206015)
南京市卫生局医学科技发展重点项目(VKX07020)
南京邮电大学校科研基金(NY206023)资助项目
关键词
新生儿疼痛
面部表情
特征提取
GABOR小波变换
核鉴别分析
neonatal pain
facial analysis expression
feature extraction
Gabor wavelet transform
kernel diseriminant
