【作者】 李春玲；

【导师】 邹北骥；

【作者基本信息】 中南大学 ， 通信与信息系统， 2008， 硕士

【摘要】 情绪是指人对认知内容的特殊态度,是以个体的愿望和需要为中介的一种心理活动。生活中人们有着各种各样的情绪,每一种情绪同时对人们的生活有反作用。因此对情绪进行识别,并相应地对情绪强度进行估计,用以指导人们的生活有着重要的现实意义。本文作者利用计算机技术,针对急躁这一负面情绪进行研究,并且将其扩展到其他情绪,具体工作如下:由于心理学家认为人的情绪通过面部表情和动作表情体现,其中的动作表情是指与情绪相关的除去面部表情以外的动作,它包括头部和身体其他部分的运动。因此本文作者将面部表情与动作表情(头部运动)二者结合来估计情绪强度。首先,对急躁情绪进行数学建模,构建了双模态特征模型,分别采用两种模态信息——人脸表情和头部运动信息来表征急躁情绪。其原因是根据实验观察,人在急躁时往往处于厌恶表情且伴有头部频繁地摆动。根据这一特点,设计用表情强度和头部摆动强度来度量情绪。表情强度采用局部特征方法计算;头部摆动强度采用头部摆动的幅度和频率进行估计,其中幅度采用情绪特征数据的方差进行计算,而频率采用情绪特征数据经DCT变换的期望值来计算。该方法能够扩展到其他情绪,为情绪强度估计提供了一种通用的特征提取方法。实验结果表明该方法能较准确的提取情绪特征。其次,针对急躁情绪强度估计这一问题,为了更好地结合面部表情与头部运动信息,本文作者采用双层隐马尔科夫模型(DMHMM)对情绪强度进行估计。第一层MHMM对面部表情进行识别,进而得到面部表情强度信息;第二层结合面部表情特征与头部运动特征进行情绪强度的估计,得到情绪强度的三个等级。实验结果表明:该算法实现简单,并且能快速的对情绪强度进行估计,识别的正确率超过了只采用面部表情信息的方法。更多还原

【Abstract】 Emotion is a special attitude at human’s cognized content, and a kind of mind which takes individual’s will and demand as resonance. There are various emotions in our living and each of them reacts to our living at the same time. As a result, it has a crucial meaning to give lives a guidance that recognizing emotion and estimating its intensity. We research on a negative emotion—impatience as well as other emotion by computer technoledge as follows:Psychologists consider that emotion is expressed by facial expression and action’s expression. Action’s expression is defined as these movements that have something to do with emotion except facial expression. Then we estimate the emotion intensity by facial expression combined with action’s expression (head movement).Firstly, we advance a mathematical modeling for impatience to build a double-mode feature model, expressing impatience with both facial and action’s expression. By observation, the reason is that disgust expression is always put on people’s face with his head waving frequently when he is impatient. According to this, we plan to measure emotion intensity by expression and head waving intensity. The expression intensity is measured by part-feature method. The amplitude of head waving is measured by variance of feature data, while the frequency by expectations concluded from the DCT changes of feature data. And the head waving intensity is estimated by the amplitude and frequence. This method can be used to estimate other emotions, which become a universal feature model. The experiment results indicate that this method can be used to extract emotional feature precisely.Secondly, we estimate emotion by Double layers of Gaussian Mixture Hidden Markov Model (DMHMM) for the estimation of impatience intensity. DMHMM recognizes the facial expression to capture its intensity in the First Layer and concludes three levels of emotional intensity by estimating the emotional intensities of facial expression’s feature and head waving’s feature in the Second Layer. The experiment results indicate that it is a simple and effective algorithm of intensity estimation and the accuracy is heightened compared with the method based on facial expression.更多还原