【作者】 王赋攀；

【导师】 戴小文；

【作者基本信息】 西南交通大学 ， 电力电子与电力传动， 2005， 硕士

【摘要】 人脸检测是确定人脸的位置、大小、数目的图像处理过程。人脸检测是人脸识别、人机交互、智能视觉监控等工作的前提。近年来,在模式识别与计算机视觉领域,人脸检测已经成为一个受到普遍重视、研究十分活跃的方向。 本文针对简单背景下的彩色单人脸图像,将肤色分割、模板匹配与区域优化工作结合起来,构建了一个基于工程实际的人脸检测系统,并对复杂背景和多人脸图像情况下的检测进行了一定探讨。 论文的主要工作包括: 首先概括的介绍了人脸检测技术研究的背景和现状,阐明人脸检测技术发展的重要意义,对目前常用的一些检测算法进行了总结。 肤色是人脸重要特征,在通过肤色采样统计和聚类分析后,确立了一种在YCbCr空间下的基于高斯模型的肤色分割方法。在YCbCr色彩空间中建立肤色分布的高斯模型,得到肤色概率似然图像,在基于Fisher评价准则的动态阈值选取算法下完成肤色区域的分割和检测。 采用数学形态学和一些先验知识对所得到的肤色区域进行优化、筛选,进一步剔除由于干扰因素造成的疑似肤色区域和非人脸肤色区域,降低了后端检测算法的复杂度。 在掌握模板匹配的原理和实现算法后,使用一种平均脸模板匹配算法进行人脸检测,平均脸模板采用手工制作的方法。实验中使用该方法进行了人脸检测,并针对非端正人脸和人脸尺寸不规则的问题,引入弹性模板的方法辅助分析。 最后,分析肤色分割和模板匹配算法在检测工作中各自的优势和缺点,将两种方法结合起来,构建了一个新的人脸检测系统。对通过肤色分割得到的候选人脸区域,使用模板匹配算法进行确认,得到人脸检测的最终结果。 实验结果表明,系统算法是有效的,具有较高的检测性能。更多还原

【Abstract】 Human face detection is to locate the faces and determine their size, position and amount in an image.As the first step of person identification, human-computer interface、 intelligent scene supervision system , face detection has been paid much attention and became a very active researeh branch in pattern recognition and computer vision application areas with its significant value in recent years.In this thesis , a face detection system based on certain project is built to detect single face in color image with simple background which integrates the skin color segmentation, face template matching and region optimization .In addition, the detection to multiple faces in color image with complex background is discussed too.The main contributions are as follows :Firstly , we introduce the face detection’s background and status in quo resumptively. The main classification of detection method is enumerated too. Human face detection problem has been intensively researched as an independent problem because of its importance in many applicative fields.Then the thesis discuss the skin’s performance in difference chrominance space in the discussion of skin color model.The YCbCr information is used to construct the Gaussian Model for skin color and get the likely region of face ,then skin color segmentation is carry out with the method of dynamic threshold optimizing according to the Fisher criterion .Some methods are used to filtrate the skin regions according to morphological and general knowledge of human face.The error regions can be eliminated and the candidate face areas can be obtained.Finally , the average face template and flexible template matching are used to verify whether the probable face area is a face indeed , and the final detection result will be obtained .The Experimental results show that the system is feasible and effective .更多还原