摘要
为了更准确地识别人脸的表情信息,采用加权主元分析识别人脸表情.首先通过小波变换进行图像分解来抽取面部区域的有效鉴别特征,然后将特征加权和主元分析相结合,根据加权重建误差最小化,计算出各类训练样本的加权子空间,最后计算测试样本到加权子空间的Mahalanobis距离,并根据距离进行分类识别.通过CMU人脸表情数据库试验证明,该方法与传统的主元分析相比可以在不增加运算量的情况下大大提高识别率.
In this paper, a new method of human facial expression recognition based on weighted PCA is proposed. At first, the wavelet transformation is used to extract the effective feature for identification on the face area. Then the weighted features are combined with PCA. After that, the weighted subspaces for each class of training sample are calculated by minimizing the weighted reconstruction errors. Finally, the Mahalanobis distances from the tested samples to the weighted subspace are computed and the classified recognition is carried out according to the distances. Based on the human facial expressing database of CMU, the experiment shows that this method can increase the recognition rate significantly without increasing the computation, when compared with the traditional PCA.
出处
《沈阳理工大学学报》
CAS
2008年第2期35-39,共5页
Journal of Shenyang Ligong University
关键词
小波变换
主元分析
加权分析
表情识别
wavelet transformation
PCA (Primary Component Analysis)
weighted features
facial expression recognition