【作者】 王浩；

【导师】 林克正；

【作者基本信息】 哈尔滨理工大学 ， 计算机应用技术， 2013， 硕士

【摘要】 人脸面部表情识别技术目前主要的应用领域包括人机交互、安全、机器人制造、医疗、通信和汽车领域等。人脸表情识别技术是涉及数字图像处理、运动跟踪、情感计算、模式识别、生理学、机器视觉、心理学、生物特征识别等领域的一个富有极具挑战性的综合交叉学科课题。人脸面部表情蕴含着丰富的情感和心理信息，能够一定程度的反映出人的大脑的思维活动。面部表情识别主要涉及两个问题，一个是怎样获得人脸面部表情的有效特征和怎样有效的分析表情特征并正确的识别。本论文主要的研究内容与创新工作包括以下内容：1．为了克服表情识别中选择移位和光照不均对识别效果的影响，提出了一种改进的局部二值模式算法。该算法具有极强的旋转不变性和灰度不变性，而且能容忍一定程度的图像旋转和不变性。对传统的局部二值模式算子进行改进,从而对于有噪声的情形下和样本图像具有低分辨率时，该算法更加具有鲁棒性和稳定性。2．针对Gabor小波特征提取后特征向量维数高的问题，提出了一种数学稀疏表示的Gabor小波表情识别算法。Gabor小波的降维是其应用的关键所在，传统的PCA算法可以降维，但是PCA没有考虑到各类特征之间的区分性。稀疏表示理论是将信号投影到变换空间上从而得到紧凑并准确的表示，因此使用稀疏表示进行降维处理，可以有利于后续的表情识别的准确率。3．针对传统的小波对于在图像边缘提取特征时有明显的不足，而图像边缘包含丰富的人脸表情信息，提出基于Curvelet特征的人脸表情识别算法。Curvelet特征能够很好的包含图像的边缘信息，有利于表示人脸表情的特征参数。在不同的尺度上Curvelet系数所包含的图像的纹理信息也不同，通过对图像进行Curvelet变换后提出去Curvelet系数，选取合适的Curvelet系数作为表情特征，可以很好的描述表情的纹理信息，识别出不同的面部表情。更多还原

【Abstract】 The technology of facial expression recognition is mainly applied in areas ofhuman-computer interaction,security,medical,communications,automotive and soon.Facial expression recognition is a topic that is full of challenginginterdisciplinary,because facial expression recognition has relation to the topicsthat included digital image processing,motion tracking,affectivecomputing,pattern recognition,physiology,machine vision,psychology andbiometrics.Expression of the human face contains a wealth of emotional andpsychological information that can reflect the thinking activity of brain.Thefacial expression recognition involves two major problems,one is how to get theexpression characteristics of the human face and the second is how to effectiveanalytical expression feature and get the correct identification.The main contentof this paper is as follows.1．In order to overcome the recognition effect expression recognition shiftand uneven illumination,an improved local binary pattern algorithm isproposed.The algorithm has strong rotational invariance and gray-scaleinvariance,and can tolerate a certain degree of image rotation and invariance.Inthis paper the traditional local binary pattern algorithm is improved,so thealgorithm is more robust and stable when the sample image has noise or a lowresolution.2．In the expression recognition the feature vector that is extracted by usinggabor wavelet has huge dimension,a method that combine gabor wavelet toextract the feature and sparse representation to reduce dimensions is proposed inthis paper.To reduce dimensions after using the gabor wavelet is the key to itsapplication.The traditional PCA algorithm can reduce the dimensions,but PCAdoes not take into account the distinction between the various types of characteristics.The theory of sparse representation is the projection of the signalto transform space and obtained a compact and accurate representation.Thereforeusing sparse representation to reduce the dimension can be beneficial to theaccuracy of the expression recognition.3．In response to the wavelet transform has obvious deficiencies whenextracting feature in the edges of the image,but image edge contains a wealth offacial expression information,so a method of facial expression recognition basedon curvelet feature is proposed in this paper.The curvelet feature contains lots ofthe edge information of the image, which is conducive to represent the feature offacial expression.In the different scales the curvelet coefficient of image containdifferent texture informations,the curvelet coefficient is extracted after using thecurvelet transform on the images and selecting the suitable curvelet coefficient asthe expression feature,which can describe the facial expressions texture very welland recognize the different facial expressions.更多还原