微表情数据库的建立和微表情检测技术研究

发布时间：2018-03-20 13:56

本文选题：微表情数据库　切入点：面部行为编码系统　出处：《山东大学》2017年硕士论文　论文类型：学位论文

【摘要】：微表情是一种不受心理控制的面部表情,具有持续时间短暂、变化幅度微弱、动作区域较少等明显区别于宏表情的特点。当前微表情在国家安全、司法审讯、谎言测试等领域潜在的应用价值引起了人们极大的关注。随着计算机模式识别技术的发展微表情相关研究取得了很多成果,但作为研究基础的微表情数据库由于微表情捕捉困难、采集过程复杂、人工编码费时耗力和图像质量评价标准缺失等原因导致数据库样本数量不足,质量参差不齐,越来越无法满足微表情研究工作。同时由于微表情自动检测技术发展滞后,无法有效的辅助复杂的人工编码,在很大程度上也制约了微表情数据库建立工作的进一步发展。针对上述问题,本文深入分析了现有微表情数据库的特点,通过合理设置实验环境,改进实验方法等措施,建立了目前样本数量最大、种类齐全、图像分辨率较高的SDU微表情数据库;随后应用不同方法从不同角度对该数据库样本进行了质量评价;最后结合微表情的表达特点,提出两种微表情自动检测方法:基于特征点集群形变矢量的微表情检测法和基于感兴趣区域光流特征矢量的幅值和角度信息的微表情检测法。本文的主要工作和创新点包括以下几个方面:第一,微表情数据库建立过程中,在微表情诱发素材选择和环境设置上获得了心理学博士的专业指导,实验器材选择视频质量各项指标相对优越的摄像器材,同时吸收心理学专业学生参与样本分析处理,最大限度地保证微表情样本的质量,最终建立了包含300个样本7种情绪类型的SDU微表情数据库。该数据库样本相对其他数据库具有分类全面、质量优良、数量最多的特点,可以为微表情检测和识别工作提供良好的实验素材。第二,针对当前微表情数据库建立标准缺失,样本质量参差不齐的现状,首次提出微表情数据库的质量评估方法。该方法包括主观评价法、客观评价法和提取特征值分析法,每种评价方法分别从分辨率、帧率和编码比特率等指标分析评价了 SDU微表情数据库质量。根据各种方法的评价结果初步得出微表情数据库质量与各指标相关性大小。第三,针对当前微表情数据库建立过程中人工编码耗时费力的现状,提出两种微表情自动检测方法,一种是基于回归树集合思想提取人脸68个特征点后将其划分到不同的特征群,通过设置合理的阈值分析对比特征群形变规律来检测微表情;一种是应用联合级联法对齐人脸并将人脸划分为几个不同感兴趣区域,统计各各感兴趣区域微表情光流特征矢量分布规律后设定阈值,再利用光流法提取样本各检测区光流特征矢量的幅值和角度信息,通过与阈值对比后检测是否发生微表情。最后两种检测方法在SDU微表情数据库和CASMEII微表情数据库上进行检测实验后均获得了良好的实验结果。
[Abstract]:Microfacial expression is a kind of facial expression which is not controlled by psychology. It has the characteristics of short duration, weak range of change, less movement area and so on, which is obviously different from macro expression. The potential application value of lie testing and other fields has attracted great attention. With the development of computer pattern recognition technology, many achievements have been made in the research of micro-expression correlation. However, the microfacial expression database, which is the basis of the research, is difficult to capture, the process of collecting is complicated, the manual coding is time-consuming and the evaluation standard of image quality is missing and so on, which leads to the lack of sample quantity and the uneven quality of the database. At the same time, because the development of automatic microfacial expression detection technology lags behind, it is unable to effectively assist the complex manual coding. To a great extent, it restricts the further development of the establishment of microfacial expression database. In view of the above problems, this paper deeply analyzes the characteristics of the existing microfacial expression database, through setting up the experimental environment reasonably, improving the experimental method, and so on. The SDU microfacial expression database with the largest sample size, complete variety and high image resolution is established. Then, the quality of the sample is evaluated from different angles by different methods. Finally, the expression characteristics of microemoji are combined. Two automatic microfacial expression detection methods are proposed: one is based on feature point cluster deformation vector and the other is based on amplitude and angle information of optical flow feature vector in the region of interest. Innovations include the following: first, In the process of establishing the microfacial expression database, the professional guidance of PhD in psychology was obtained in the selection of microfacial expression inducing material and the setting of environment, and the experimental equipment was used to select the camera equipment with relatively superior video quality indexes. At the same time, students majoring in psychology are involved in sample analysis to maximize the quality of microfacial expression samples. Finally, a SDU microemoji database containing 300 samples and 7 emotion types was established, which has the characteristics of comprehensive classification, excellent quality and maximum quantity compared with other databases. It can provide good experimental material for microfacial expression detection and recognition. Second, aiming at the current situation of the lack of standards and uneven sample quality in the microfacial expression database, The quality evaluation method of microfacial expression database is proposed for the first time. The method includes subjective evaluation method, objective evaluation method and eigenvalue analysis method. The quality of SDU microfacial expression database is analyzed and evaluated by frame rate and coding bit rate. According to the evaluation results of various methods, the correlation between the quality of microfacial expression database and each index is preliminarily obtained. In view of the time-consuming and laborious manual coding in the establishment of microfacial expression database, two automatic microfacial expression detection methods are proposed. One is to extract 68 facial feature points based on the idea of regression tree set and divide them into different feature groups. The microexpressions are detected by setting a reasonable threshold to analyze and compare the deformation of feature groups. One is to align the faces and divide them into several different regions of interest by using the joint cascade method. The characteristic vector distribution of micro-expression optical flow in each region of interest is analyzed and the threshold is set. Then the amplitude and angle information of the characteristic vector of optical flow are extracted by optical flow method. The last two methods were tested on SDU microfacial expression database and CASMEII microfacial expression database, and good experimental results were obtained.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】