基于bayesian网络的面部情感判别分析研究

项目名称： 基于bayesian网络的面部情感判别分析研究

项目编号： No.60805007

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 轻工业、手工业

项目作者： 何良华

作者单位： 同济大学

项目金额： 20万元

中文摘要： 情感占据了人类精神世界的核心地位，对情感的直接表现形式-人脸的分析与研究不仅有助于理解情感，而且将极大地促进人工智能、人机交互等学科的发展。目前研究人员从表情的几何结构和统计特性两个角度对表情识别进行了研究，提出了很多经典算法，但是算法的易实现性和模型可解释性之间的矛盾一直没有得到有效地解决。本项目从人脸特征提取开始，一直到Bayesian网络模型构建，研究了一整套高效，高精度的人脸分析方法和模型，具体而言，提出了基于Gabor小波与局部二元模式的人脸特征的提取算法，提出了基于多维主分量、联合核主分量与共同向量的人脸统计模型。另外，还研究了基于稀疏数据的LLE改进模型，取得一系列成果。研究Bayes统计理论和图的直观表达来研究表情的统计特性和特征间的逻辑关系，对贝叶斯网络结构进行理论分析，针对经典算法不足以完全开发贝叶斯网络的图形特性导致结构学习的低效和较低精确性，研究了基于马尔科夫集的贝叶斯图形增强结构学习算法，已取得长足进展。基于贝叶斯网络的表情识别研究，对于分析表情、提高表情识别率、理解情感具有重要的理论意义和实用价值。

中文关键词： 人脸识别;贝叶斯网络;主分量分析;局部线性嵌入法;局部二元模式

英文摘要： Emotion is the key point of spirit for our human being. As the direct expressing emotion organ, face is very important in analysis of automatic computer intelligence.The study on face image is not only benefit to face expressing udnerstanding, but also benefit to improve the related areas research, such as artifical intelligence,man-machine interaction etc. Researches have proposed a lot of classical algorithms form face geometry and statistical aspect. but the contradiction between implement and interpretablity of algorithms has not been effectively solved. This project had made a series studies for efficient and high-precision face analysis. from the beginning of facial feature extraction to the facial feature analysis using Bayesian network model. In detailed, we proposed a facial feature extraction method based on Gabor wavelet and local binary pattern extraction. We also study the facial feature statistical properties using multi-dimensional principal component, the combined method of kernel principal component and vommon vector. We still studied the LLE improved model based on sparse data and made a series of achievements. We also studied the bayesian network method for facial expression analysis. Expescially, we analyzed the Bayesian network structure theoretically using markov banket. Some significant achievements have bee acquried. Expression recognition based on Bayesian networks is very meaningful for analyzing of expression,improving the achievement of facial expression recognition, understanding emotions.

英文关键词： face recognition; bayesian network; principal component analysis;local linear embedding; Local binary pattern