【作者】 马飞；

【导师】 冯乔生； 夏幼明；

【作者基本信息】 云南师范大学 ， 计算机软件与理论， 2006， 硕士

【摘要】 人脸表情识别有着广泛的应用前景，逐渐成为当前人机交互领域的研究热点之一。表情识别是智能化和自然的人机交互的本质。实现自然和谐的人机交互，必须使计算机能够有效地理解人的情感和意图。目前与表情分析识别相关的研究主要集中在生理学、心理学和认知科学等领域。迫切需要开展对人脸表情的自动分类方法的研究，这项研究不仅对人机交互有十分重要的意义，而且在计算机辅助的训练及远程教育等众多领域都具有潜在的应用价值。 本文的研究工作是基于人脸的几何特征进行眼睛、眉毛定位以及表情分析与识别的，主要进行了以下五个方面的研究，取得了一定的研究成果。 (1) 在人脸表情原图像的预处理过程中，为防止出现一些小的断点产生不利的影响进行了一定断点连接，提出了基于连通区域的探针法进行上边缘的提取，在对上边缘进行编号记录时对邻接点的范围进行了扩展，提出了扩展六邻域的方法来寻找同一边缘上的点，实验表明本文所提出的方法有效地对断点进行了连接。 (2) 本文结合眼睛、眉毛之间的几何特征约束，提出了基于上边缘相邻四端点的方法进行眉毛和眼睛上边缘的精确定位。 (3) 本文利用以下几个特征对人脸表情的变化进行刻画，眉毛、眼睛的两端点间的斜率、中点到两端点所在直线的距离、嘴巴的上下嘴唇与嘴角间的张角以及上下嘴唇的距离与左右嘴角的距离之间的比值，本文对以上特征数据进行了提取。 (4) 本文提出了一种新的多维特征数据权重函数用于构造各维表情数据的权值，该函数对人脸表情的各维特征数据进行了离散化，增加它们的区分度，此结果已用于人脸表情识别系统中。 (5) 本文在上述研究的基础上，使用VC++6.0完成了人脸表情识别系统FERS，实验表明本文提出算法的有效性及实时性。更多还原

【Abstract】 Facial expression recognition that has the extensive application future is gradually the hotspot of the human-computer interaction. Facial expression recognition is the essence of the intelligence and nature human-computer interaction. To realize the nature and harmonious human-computer interaction must make the computer understand the emotion and intention. At present, the research on the facial expression mainly focuses on the physiology, psychology and cognitive science etc. It is important to develop the automatic classification of facial expression, ant not only the human-computer interaction is important, but also there are many applications in the field of computer assistant training and the network education.The research of this paper is mainly based on the facial geometrical feature to locate the brow, eye and the facial expression analysis and recognition, there are 5 parts mainly, and there gets some research results in the process.(1) In the process of face original image pretreatment, to connect the break is to avoid the disadvantage of some small break points, this paper presents the up edge extraction method of the probe based on the connected region, and extend the neighbor pixel range while coding the up edge pixels, and this paper presents the extended 6-neighborhood to store and code the edge, the experiments show that the algorithms are effective.(2) This paper presents the method of up-edge four neighborhood endpoints with the geometrical feature restriction of the brow and eye to locate the brow and eye with the geometrical feature restriction of the brow and eye;(3) This paper depicts the change of facial expression with the following features: the slope of the two endpoints of the brow and eye, the Euclidean distance from the middle point to the line of the two endpoints, the opening of up-down lips and the corners of the mouth, the ratio of the distance of up-down lips and left-right corner.(4) This paper presents a new function of every dimension data to calculate the weight of the expression feature data, this function disperses every dimension data possibly, improve the differentiability of the data, which is better in practice.This paper develops the facial expression recognition system (FERS) in VC++6.0 based on these researches, the experiment shows that the algorithms presented by this paper are effective and the system has the characteristic of the real-time.更多还原