【作者】 辛威；

【导师】 朱虹；

【作者基本信息】 西安理工大学 ， 模式识别与智能系统， 2004， 硕士

【摘要】 近一段时期，在图像分析和模式识别领域的发展使感情和会话时的脸部信号的自动识别成为可能。自动化后的人脸表情识别可以被应用到人机交互方面，从而最为一种新的方式和手段从而使人机之间的交流更紧密，更高效。 本文主要针对人脸表情自动识别方法进行了研究。所提出的人脸表情的自动识别方法主要分为三步，首先是人脸检测，文中提出了一种基于色彩空间转换和亮度补偿的肤色检测算法，然后通过构建眼，嘴模板来进行人脸的特征点定位，以及获得图像中人脸所在区域。在定位人脸后，下一步就是要进行脸部表情信息的自动提取，为了描述人脸的面部结构，本文采用了Gabor滤波器对图像进行滤波，由于Gabor矢量在相邻像素间是高度相关和信息冗余的，本文构建了一个脸部稀疏网格，抽取网格中每个节点处的一组Gabor滤波结果作为脸部表情数据，从而实现脸部表情数据的自动提取。获得所需信息后，最后一步就是要进行表情的分类，根据所使用的人脸数据库，将人脸表情分为三类：自然，高兴，生气。分类器的设计采用的是基于自组织神经网络的方法，为了克服传统的自组织映射神经网络的训练结果容易受训练样本的输入顺序和权值初值影响，而导致训练结果不符合期望的问题，因此，在训练过程中引入了监督机制，以使训练结果与期望相符。通过以上三步就完成了对人脸表情的自动分类。 本论文的研究主要利用了图像处理方法和模式识别知识对人脸表情进行自动分类，实现了一个人脸表情自动分析系统。实验结果表明，该系统能够达到很好的分类效果。更多还原

【Abstract】 Recent advances in image analysis and pattern recognition open up the possibility of automatic detection and classification of emotional and conversational facial signals. Automating facial expression analysis could bring facial expressions into man-machine interaction as a new modality and make the interaction tighter and more efficient.Our aim is to explore the issues in design and implementation of a system that could perform automated facial expression analysis. In general, three main steps can be distinguished in tackling the problem. First, before a facial expression can be analyzed, the face must be detected in a scene. Based on color transformation and a novel lighting compensation technique, skin regions can be detected over the entire image; then eye and mouth maps are constructed for verifying each face candidates. The second step is to devise mechanisms for extracting the facial expression information from the observed facial image. Images are transformed using a multi-scale, multi-orientation set of Gabor filters. Since Gabor vectors at neighboring pixels are highly correlated and redundant, a rectangular grid is then automatically registered with the face based on the result of face detection. The amplitude of the complex valued Gabor transform coefficients are sampled on the grid and combined into a single vector. After extracting facial expression information, the final step is to define some set of categories, which are used for facial expression classification and /or facial expression interpretation, and to devise the mechanism of categorization. According to the utilized face database, three facial expression categories are defined: neutral, happiness and anger. The categorization architecture is based on a SOM. In order to eliminate influence of initial values and sequence of input examples in SOM, supervised learning is introduced into the training stage. By three steps mentioned above, theautomated classification of facial expression is realized.In this paper, an automatically facial expression analysis system is constructed based on the image processing and pattern recognition. Experiments show good classification result can be obtained through our system.更多还原