摘要
针对目前表情识别类间信息无关状态,提出了一种表情类间学习的神经网络分类识别算法。该算法首先构建一个BP网络学习对和一个距离判据单元,该距离判据单元仅用来计算类间的实际距离,类间期望距离是根据大量实验结果获得的;然后通过类内实际输出和类间期望距离来修正该网络;最后给出一组实例样本进行表情分类识别。实验结果表明,该算法能有效地识别人脸表情,能紧密地将各类表情间的信息联系起来,效率和准确性均有明显提高。
In view of the current unrelated status of expressions recognition in congener information, this paper proposed an algorithm of the neural network classification of expression congener learning. The algorithm first built a pair of network of BP and an unit of the distance judgment evidence which was only used to calculate the actual distance between categories, and the congener expected distance was obtained under a lot of experiments, then amended the network by the actual output of inner class and the expected distance of congener, finally tested expression recognition through a set of samples. The experimental results demonstrate the feasibility of the algorithm which is able to link information between two expressions at random closely. It shows that the algorithm can improve the efficiency and accuracy of facial expression recognition obviously.
出处
《计算机应用研究》
CSCD
北大核心
2008年第7期2219-2222,共4页
Application Research of Computers
基金
湖南省自然科学基金资助项目(06JJ50109)
关键词
表情识别
类间学习
神经网络
类间期望
距离判据
expression recognition
congener learning
neural network
congener expectations
distance judgment evidence
