智慧学习环境中基于面部表情的情感分析

摘要

情感与认知加工之间存在着密不可分的联系，学习过程中的情感状态对学习效果有一定的影响。在智慧学习环境中实现学习者情感分析，有利于促进智慧学习的发生。表情作为人类情感表达的主要方式，其中蕴含了大量有关内心情感变化的信息，通过面部表情人们可以推断内心微妙的情感状态。目前，人脸检测技术已经实现了从复杂背景中定位人脸，分类算法也相对成熟，因此表情识别的研究工作主要集中在表情特征提取上，而现有研究基本上都是基于人脸与表情的混合特征进行的识别，这产生了较大的干扰。在表情识别时，理想情况是将个体相关的人脸特征和与个体无关的表情特征相分离。依据心理学家Ekman提出的FACS（面部表情编码系统）构建的智慧学习环境下基于面部表情识别的情感分析框架，通过特征分解将个体特征及表情特征分解到不同的子空间，在表情子空间中进行表情识别，从而排除个体特征对表情识别的干扰。经JAFFE表情库的验证，表情识别结果比较理想，已在三维虚拟学习平台Magic Learning的师生情感交互子系统上实现了基于面部表情的学习者情感识别及情感干预。%Emotion has many relative connections with cognition. Affective states in learning have certain influence on learning effect. Learners' emotion analysis in smart learning environment is beneficial to smart learning. At present, locating human face in complex background has already been actualized by face detective technology, and some available algorithms have also been applied to expression classification. Hence, the research into expression feature extraction has become the major issue in the expression recognition area. The feature extracted by existing methods is the combination of individual's facial feature and expressional feature, which greatly obstruct the expression recognition. In an optimal situation, the related individual facial feature can be separated during the process of expression recognition. According to FACS (Facial Action Coding System) proposed by Ekman, a famous psychologist, we constructed an emotion analysis framework based on facial expression recognition in smart learning environment. We used feature decomposition method to decompose the facial feature and the expressional feature into face subspace and expression subspace respectively. After that, expression recognition will be done in the expression subspace and the interference of facial features will be eliminated. Experimental results on JAFFE database suggest that our method is effective. Facial expression recognition for emotional intervention has been performed in Magic Learning, which is an emotional interaction subsystem between learners and virtual teachers in 3D virtual learning environment.