【作者】 张丹丹；

【导师】 彭延军；

【作者基本信息】 山东科技大学 ， 计算机应用技术， 2008， 硕士

【摘要】 由于人脸检测识别技术在安全部门、电视会议、身份鉴别、数字监控等领域的应用越来越广泛,它作为一项极具发展潜力的生物特征识别技术,已成为近年来科学研究的热点,也已引起了研究者们越来越多的关注。本文针对彩色图像中人脸的检测和识别技术进行了详细的研究,提出了人眼中心定位和简化人脸识别的新方法,并对现有算法进行了有效的改进。本文主要包括两部分的内容,一部分是人脸检测,另一部分是人脸识别。将人脸检测分为粗定位和精确定位两部分。在讨论及分析肤色模型概念及原理的基础上给出了基于肤色模型的人脸分割方法,从而实现了人脸的粗定位。在Canny算子边缘检测的基础上,提出了一种人眼中心的定位方法,在确定眼睛大致区域后,利用圆的几何特性在边缘图像中寻找眼睛中心点。综合运用亮度和色度信息定位鼻子和嘴巴中心点以及其它特征点,并将三次B样条曲线用于鼻子轮廓的拟合。最终实现了人脸的精确定位,排除了脖子、服饰等干扰因素。在人脸识别部分,基于隐马尔可夫模型,提出了一种简化观察序列提取过程的方法。通过小波变换压缩原始图像的数据,利用人脸检测阶段所得到的结果缩小采样范围,定义恰当尺寸的采样窗口,直接以人脸采样区域的灰度均值作为特征,形成一系列观察序列,从而简化人脸隐马尔可夫模型的训练过程,减少识别人脸的时间。通过大量实验验证本文所提出的人脸检测及人脸识别方法的准确性和可行性。更多还原

【Abstract】 Human face detection and recognition has applied more and more on the area such as security system, human ID, digital surveillance and so on. As a biologic technology which possesses great developable potential, it has become one of the most active research topics, and more and more investigators put their attention on this topic.The human face detection and recognition of color image is the studied object. In this paper some new methods have been presented, and also improve on some existed methods. This paper has two main parts, one is the face detection, and the other is the face recognition.In the part of face detection, the conception and principle of skin color model have been introduced first. Then the skin color area can be segmented and the approximate face region can be located according to the skin color model. After comprising the edge detection operators which can be used on face detection, the Canny operator is chose to detect the edge information. One eyes’ positioning method is introduce. The principle is detecting circles in the eyes regions with the geometry property of circle. Then the methods of positioning the nose and mouth are introduced which used the luminance and chroma information. The 3-order based spline curve is use to simulate the border of nose. Then the exacted region of face can be located, which eliminates the factors such as neck.In this paper, the face recognition implementation is based on the HMM.A predigest method of observation sequence is proposed. Using the wavelet transform to compress the information of image. Define seemliness sampling window to extract gray as the feature in the compress region based on the results of face detection. And then the process of training HMM can be predigested.The lots of experimental results prove the validity and accurately of the algorithms of face detection and face recognition presented in this paper.更多还原