基于面部遮挡的大学生心理测试视觉情绪计算模型研究

项目名称： 基于面部遮挡的大学生心理测试视觉情绪计算模型研究

项目编号： No.61300119

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 王晓华

作者单位： 合肥工业大学

项目金额： 23万元

中文摘要： 针对现有大学生心理健康测试过于依赖问卷，而问卷一定程度上却不能反映真实测试情绪的问题，探究并提出一种基于视觉的情绪计算模型，并对由此引出的科学问题及关键技术开展深入系统的研究。课题首先突破目前大多人脸表情识别系统难以摆脱光照、背景、部分遮挡的局限，瞄准从解决遮挡入手提高表情识别的鲁棒性；而针对面部表情遮挡，采用基于RPCA和人脸中心线对称相结合的方法，创建算法的自适应调整机制，做到对遮挡合理检测并有效重构。然后在此基础上引入遮挡面部的手势识别，走出表情识别过程中把手势信息当作噪声盲目去除的误区。为有效解决面部表情/手势类别与情绪目标对应时的不确定性，本课题借助动态Bayes网络对模型的计算推理开展研究，并有计划、有步骤地应用到拟搭建的自发表情视频数据库、以及大学生心理健康测试的具体实践中，以稳健提高基于面部遮挡表情或面部表情/手势双模态的情绪推理能力，实现大学生心理测试时情绪的正确识别。

中文关键词： 大学生心理健康测试；面部表情；面部手势姿态双模态；视觉情感数据库；情绪

英文摘要： Mental Health test in undergraduate is presently too dependent on questionnaire,but sometimes the questionnaire could not give a real emotion of the test.Therefore,a visual computational model of emotion is explored and proposed by the project,and in-depth studies on the corresponding scientific issues and key technologies are carried out.The limitations of light,background,image occluded partially in current facial expression recognition system are not broken through,the robustness of expression recognition ability could be improved by solving the occluded images' recognition in the subject.The method used to deal with occluded images is adopted based on algorithms of RPCA(Robust Principal Component Analysis)and symmetry transformation about facial midline.The parameters could be adjusted adaptively to detect facial occlusion and make the occlusion reconstructed effectively.Afterwards,gesture expression is introduced instead of blindly removed as noise from expression recognition.Because of many uncertainties between face expression/gesture feature and emotional condition, dynamic Bayes network is used in the subject to solve the calculation problem of emotional condition, and scientific researches are made on the visual emotion model can be applied to spontaneous expression database will be built,as well as th

英文关键词： Mental health test in undergraduate；Facial expression；Facial and gestural dual-modality；Emotional visual database；Emotion