【作者】 李倩；

【导师】 黄晓生；

【作者基本信息】 华东交通大学 ， 计算机技术， 2011， 硕士

【摘要】 表情识别正逐渐成为人工情感、人工心理、模式识别等领域的研究热点。通过运用表情识别技术使计算机具有拟人的表情识别功能,能像人一样通过表情进行交流,能通过表情读懂人的内心世界,这对构建自然和谐的人机交互环境具有重要意义。表情识别技术是情感计算的内容之一,是心理学、生理学、图像处理、机器视觉、模式识别等领域的一个极富挑战性的交叉课题。本文首先在认真阅读国内外大量文献和书籍的基础上,对人脸表情识别课题背景、研究价值、应用领域及国内外研究现状进行了阐述。并对人脸表情识别系统所涉及到的各个步骤现今所采用的主流的方法做了比较全面的综述。这为表情识别技术的研究和探讨打下了坚实的理论基础。其次,对一幅要进行表情识别的表情图像都进行图像预处理工作。首先对人脸检测技术进行了深入的研究,根据人脸肤色在YCbCr空间的概率分布特性和对应的肤色分布的高斯模型特征,本文采用一种基于肤色高斯模型的人脸检测算法。在YCbCr色彩空间建立高斯肤色模型,运用这一模型对肤色进行分割,然后运用人脸在几何规则上的一些先验知识对人脸区域进行定位。图像预处理的另一个工作是对图像进行归一化。归一化的内容是利用图像灰度化、旋转变换、缩放变换、灰度均衡化这四种算法,将人脸检测得到的表情图像归一化。然后,利用Gabor小波提取表情特征的优势及弹性模板匹配方法对表情分类的优势,探索了基于Gabor小波和弹性模板匹配的表情识别方法。运用多空间尺度、多方向的Gabor滤波器组来提取表情图像的Gabor特征。并结合弹性模板匹配,计算待测表情图像与表情数据库中的表情图像的相似度,然后利用这一相似度,运用K近邻表情分类策略对表情进行分类识别。通过实验证明了该方法的可行性。最后,运用本文所介绍的表情识别各算法,使用Visual Studio C++ 6.0开发工具在Windows XP平台上设计并实现了人脸表情识别系统。经过两个表情数据库的测试,表情识别系统的识别性能基本达到了预期要求。更多还原

【Abstract】 Facial expression recognition gradually becomes a hot topic among artificial emotion, artificial psychology and pattern recognition. The purpose of facial expression recognition is to let the computer have the ability: communicating through facial expressions. If the computer can recognize human facial expression, it is easy to build a natural and harmonious environment for human-computer interaction. Facial expression recognition has the potential market value and wide application prospect. Facial expression recognition is one of contents of affective computing, and it is a challenging topic in psychology, physiology, image processing, machine vision, pattern recognition and other areas.In this paper, based on the theories of digital image processing and facial recognition, this paper integrates the face detection, image preprocessing, feature extraction of facial expression, expression classification and recognition into facial expression recognition system.Firstly, in this paper lots of literatures and books which come from home or abroad has been read during preliminary preparatory stage, on the basis of these references, this paper discuss the background, applied values and research situation about facial expression recognition. Then, this paper introduces the main existing expression recognition algorithms at home and abroad which involve in the mainstream of facial expression recognition.Secondly, the expression images should be preprocessed before expression recognition. Studying face detection technology deeply, based on the probability distribution of skin color in YCbCr color space, this paper apply the face detection algorithm based on the Gaussian model of skin color. We build the Gaussian model of skin color in the YCbCr color space, and use the model to detect face. Another work of image preprocessing is image normalization. This work involves grayscale equalization, rotation transformation, sacling transformation and image grayscale.Thirdly, on the basis of the advantages of Gabor wavelet and flexible template matching method on the facial expression recognition, we explore the expression recognition method based on Gabor wavelet and flexible template matching. At the same time, based on face detection, a family of Gabor filters which involves multi-scale and multi-orientation is used to extract the features of expression. And then Euclidean distance and KNN strategy are used to recognize and classify the facial expression. Experimental results demonstrate the feasibility of the method.Finally, this paper adopt the algorithms which are described above to design and develop a facial expression recognition system. The development tool this paper use is Visual Studio C++ 6.0 development tool in the Windows XP platform. According to the experimental results on two expression databases, the performance of facial expression recognition system has basically reached the expected requirements.更多还原