摘要
几何归一化处理对于表情特征信息的有效提取具有重要意义,常用的归一化方法由于选择的基准特征在表情变化时存在不稳定性,容易造成归一化后的结果存在误差。本文结合光流特征提取方法分析了归一化误差对表情特征提取的严重影响,并提出了加权优化匹配算法进行误差的矫正。算法以传统模板匹配原理为基础,根据各像素点的运动剧烈程度分配相应的匹配权值。实验证明,误差得到了有效的矫正,提取的表情特征信息更加真实。
Generally the geometrical normalization process is important to the extraction of facial expression features. In commonly used normalization methods, the unstability of the fiducial features resulted from facial motions always causes normalization errors. Making use of the optical flow method, the paper analyzes quantitatively the serious influence on feature extraction caused by normalization errors,and proposes a weighted optimal matching algorithm to solve it. The algorithm is founded on the basis of the template matching theory, and assigns corresponding weights to the points by their motive degrees when carrying on the calculation of correlative coefficients. Experimental results show that the algorithm can restrain the disturbance that normalization error produces,and facial expression features can be extracted more accurately.
出处
《计算机工程与科学》
CSCD
2006年第6期62-65,共4页
Computer Engineering & Science
基金
国家973计划资助项目(2004CB719404)
关键词
表情特征提取
归一化
误差矫正
光流
human facial feature extraction
normalization
error-correction
optical flow
