摘要
本文讨论了基于光流法的面部表情识别。由于面部表情运动是一非刚体运动,容易产生形变,因此标准光流法估计不准确。为此,本文通过引入div-curl样条函数作为扩展光流约束方程的附加约束条件,推导了非刚体光流算法,并给出了一阶和二阶div-curl样条约束下光流的数值解。最后将该算法用于面部表情特征提取,构建了HMM与BP神经网络混合分类器。实验结果表明面部表情识别率得到显著提高。
In this paper we address the problem of estimating the non-rigid motion in facial expression sequences. Due to the great deal Of temporal distortions that luminance patterns exhibit in facial expression images, standard optical flow algorithms are not well adapted in this context. To cope with the problem, a novel approach for estimating facial expression motion based on first-order and second-order div-curl splines constraint is presented. The numerical resolutions of this method are induced. Facial expression feature vector flows are extracted by improved optical flow algorithm and a hybrid classificer based on HMM and BP neural network is designed. The experiment results show that the performance of this approach is better than normal method.
出处
《计算机科学》
CSCD
北大核心
2007年第3期213-215,229,共4页
Computer Science
基金
国家自然科学基金项目(No.60573059)
江西省教育厅科技计划项目(赣教技字[2005]145号)
北京科技大学重点基金项目资助
作者简介
杨国亮博士生，主要研究方向：面部表情识别、情感计算与人工心理；
王志良博士，教授，博士生导师，主要研究领域为人工心理、情感计算；
王国江博士生，主要研究领域为人工心理、情感计算；
陈锋军博士生，讲师，主要研究领域为图像处理、面部表情识别。
