摘要
分析了用活动外观模板(AAM)提取的人脸表情特征来进行人脸表情识别(FER)的可行性,尝试了以此特征向量为基础的FER。根据人脸图像的特点,先用特征眼的方法定位眼睛区域,再采用AAM的优化算法获取新对象的特征,缩短了AAM方法定位新对象的优化时间,提高了定位的准确度。采用秩相关分析和非度量多维标度(nMDS)等多变量统计学方法分析得出AAM方法提取的表情特征能够很好地表达表情的变化,并构造了神经网络分类器对人脸表情图像进行识别实验,得到93.5%的识别率。
Feasibility of facial expression recognition (PER) using facial expression feature extracted was investigated by active appearance models method (AAM). According to characteristics of face images, eye region is extracted using eigen-eye method firstly to decrease AAM optimization time and improve accuracy of locating new objects using AAM. Rank correlation method and non-metric multidimensional scaling (nMDS) method are adopted to analyze the performance of facial expression feature extracted by using AAM method. It is proved that this kind of feature has a good ability to represent expression variability. An artificial neural network was constructed to recognize facial expression. Experiments show that the recognition rate is 93.5%.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2004年第7期853-857,共5页
Journal of Optoelectronics·Laser
