基于多层潜式条件随机场的视频情感事件检测研究

项目名称： 基于多层潜式条件随机场的视频情感事件检测研究

项目编号： No.61272206

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 余新国

作者单位： 华中师范大学

项目金额： 81万元

中文摘要： 视频情感事件检测是计算机领域中众多学者关注的热点研究问题，原因在于它是多个领域急需的关键技术。本项目旨在为视频情感事件检测创建更有推理能力的新模型，并建立能提取有区分力特征的新方法。具体而言，主要拟展开以下三方面研究：一、克服单层条件随机场不能实现多层推理的不足，提出多层潜式条件随机场，并创立基于这种新模型的情感事件检测方法。二、面向情感事件检测改进基于局部二值化模式的脸部表情识别方法。其中，为克服传统脸部表情识别方法不能利用特征间关联信息的不足，提出基于潜式条件随机场的脸部表情识别方法；为克服局部二值化模式在描述脸部动作方面的不足，提出将局部二值化模式与脸部特征信息联合起来作为脸部表情识别的特征。三、研究如何用改进的边缘直方图和谱系回归识别脸部特征及状态。本项目预期为视频情感事件检测创造新理论模型、新特征提取法和有效的新检测算法，这将为开发具有知识产权的智能情感交互产品打下坚实基础。

中文关键词： 视频情感事件检测；条件随机场；随机森林；情感学习状态检测；机器智能解答

英文摘要： Visual emotion event detection (VEED) has been being a hot research topic and many researchers have been paying a lot of attention to this problem due to that it is a key technology demanded by multiple relevant research areas. This project aims to propose the new variants of the latent conditional random field (LCRF) and the new method of extracting the discriminative facial expression features for VEED. Particularly, we plan to do research in the following three aspects. First, to overcome the drawback of the existing single-layer LCRFs that they cannot do the multi-layer inference this project proposes a multi-layer latent conditional random field model (MLCRF) for VEED being more complicated than visual hehavior detection. Second, to explore a better emotion detection method this project aims to enhance the facial expression recognition. It firstly adopts the modified LCRF to recognize the facial expression. This would fix the drawback that the exiting methods do not use the temporal relation of features. It also joins the information of the facial feature with the local binary pattern of face and conducts facial expression recognition on the joint feature. Third, to acquire the accurate information of the facial features this project will use the circular histogram of oriented gradients to do facial featur

英文关键词： Video Emotion Event Detection；Conditional Random Field；random forest；affect-sensitive e-learning；automated solution