摘要
将Ｅｉｇｅｎｆａｃｅ多子空间分类方法用于面部表情识别；针对传统多子空间分类方法中的问题和缺点，提出了两种变维分类方法——静态变维分类和动态变维分类．根据脸部不同区域所含表情成分的不同，将人脸图像划分成表情子区域，构成表情子图像；并分别对各类表情子图像集求解其表情特征子空间．在识别时，用变维分类方法把表情子图像分别投影到各个表情特征子空间上，根据该图像与其在表情特征子空间的投影之间的相似性来进行分类．
In this paper, multi sub space classification method of Eigenface is applied to human face expression recognition. To overcome the disadvantages of traditional multi sub space classification which is not very precise in describing respective expression patterns, the paper presents two approaches of dimension variant classification——Static Dimension Variant Classification and Dynamic Dimension Variant Classification. Since different human face's regions contain different expression ingredients, an expression region extracted from a human face image is divided into expression sub regions to form expression sub image (expression pattern vector), and the eigen sub space of each kind of expression is worked out using a group of training sub images. In recognition, an unknown human expression sub image is projected onto a group of eigen sub spaces whose dimension adopted are different and can be adjusted according to this unknown expression sub image, and the likelihoods(eigen distance) between the expression sub image and it's projections on all the eigen sub spaces are applied to classification and recognition. Static Dimension Variant Classification method overcomes the inability of traditional multi sub space classification in describing expressions' distributing features when those expressions' distribution is sophisticated, and gives a much more well description of expressions' distribution. On the basis of Static Dimension Variant Classification, this paper puts forward Dynamic Dimension Variant Classification which makes use of specific samples' location information and decreases the distributing disorder of different expression classes, makes the description of expression more precise and contributes to the recognition ratio.
出处
《计算机学报》
EI
CSCD
北大核心
1999年第6期627-632,共6页
Chinese Journal of Computers
基金
国家八六三高技术研究发展计划
中国教育部跨世纪人才计划基金
关键词
EIGENFACE
变维分类
表情识别
人脸识别
Eigenface, dimension variant classification, expression recognition, face recognition.
