融合认知机理的概率图模型表情识别方法研究

项目名称： 融合认知机理的概率图模型表情识别方法研究

项目编号： No.61503277

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 刘志磊

作者单位： 天津大学

项目金额： 18万元

中文摘要： 面部表情是人类进行情感表达和推理最直观的途径。近年来，研究人员主要在两个不同方面对表情进行了广泛而深入的研究：一方面是基于认知科学的表情认知研究，另一方面是基于计算机科学的表情识别研究。然而，表情认知研究中所揭示的表情认知机理却对表情识别研究中识别模型的设计和训练具有重要的参考意义。本研究试图通过设计不同条件下的表情认知实验，采集受试脑部的fMRI数据，进一步从不同方面进行分析揭示人类对表情认知的机理。鉴于概率图模型在不确定推理、变量关系建模等方面的优势，本研究将基于概率图模型进行表情识别方法的研究，并将所得到的表情认知机理作为隐性知识辅助表情识别模型中结构和参数学习，最终根据训练得到的模型参数以及识别效果对认知机理在表情识别中的作用进行验证和分析。

中文关键词： 表情识别；表情认知；概率图模型；脑科学；认知机理

英文摘要： Facial expression is the most intuitive way for human emotion expression and perception. Recent years have seen extensive research on facial expression are conducted on two different fields: one is expression cognition research based on Cognitive Science, and the other is expression recognition research base on Computer Science. However, some cognitive mechanisms revealed trough expression cognition research have important significance for reference in model design and learning in expression recognition research. In this study, the fMRI data of human brain during different facial expression cognition tasks will be collected, and some expression cognitive mechanisms will be explored through the analyses from different aspects. In light of the powerful abilities of Probabilistic Graphic Model on uncertainty inference and variable relation modeling, in this study, the facial expression methods will be studied based on probabilistic graphical models, and the cognitive mechanisms obtained during our expression cognition research will be adopted as hidden knowledge for structure learning and parameter learning. Finally, the effectiveness of the cognitive mechanisms will be verified and analyzed through the trained model parameters and expression recognition performance.

英文关键词： Expression Recognition;Expression Cognition;Probabilistic Graphical Model;Brain Science;Cognitive Mechanism