【作者】 徐庆；

【导师】 石跃祥；

【作者基本信息】 湘潭大学 ， 信号与信息处理， 2008， 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。随着计算机应用的普及、性能提高以及图像和模式识别领域的研究逐步成熟,对人脸检测的研究也将受到越来越多的重视。虽然近几十年来出现了大量的人脸检测的算法,但检测率、检测速度以及算法鲁棒性等方面还是不够理想。因此本文对人脸检测进行了深入的研究,主要工作在如下三个方面:1.本文在灰度空间提出了一种利用眼睛和鼻子的灰度特征和几何特征的人脸检测方法。选取眼睛和鼻子作为特征点,构造一个三角的特征模型。另外,此方法对候选特征图像采用逐步改变分块大小的方法进行搜索,得到独立的特征点,并利用人脸结构特点的先验知识建立模型的搜索策略。实验表明,此方法能迅速准确的从复杂背景中检测出人脸,而且对多人脸同样有效。2.根据肤色色度的分配比例,本文提出了一种改进YCbCr的肤色检测方法,用Cg分量代替Cb分量,实验表明在改进的颜色空间内,肤色投影到CrCg平面内了更好得聚类效果。3.在肤色背景下,利用肤色和唇色在YUV空间分布特点,变换YUV空间的坐标轴,增大唇色和肤色V分量上的差异。提取的唇色的质心和旋转方向,根据人脸几何特征的先验知识建立人脸定位模型。实验表明唇色定位人脸算法简单,速度快,更具实效性,对旋转的人脸同样有效。更多还原

【Abstract】 Face detection is an image processing course, ascertaining face’s location, size and amount in an image. And face detection is the precondition of face recognition, human-computer interface and intelligent scene supervision system. Research on face detection will be paid more attention, with computer application being popular and its capabilities improved and researches growing up in image processing and mode recognition field. There are a lot of arithmetics of face detection in the past decades. But the rate, precision and robustness of arthmetics are not ideal. The research works in this paper is as following.The key problem of face detecting re is extracting features in gray space.. In this paper, we proposed a method of Feature Model Based Face Detecting. It extracts features from eyes and nose, and then constructs a triangular feature mold component with eyes and nose. The search is taken in the candidate feature image by changing the size of blocks step by step. This processing can obtain separate points of feature mold. And finally proposed a strategy based on the knowledge about facial structure/distribution to search the feature mold. Experimental results prove that this method can rapidly and accurately detect face in an image with complex background. Furthermore, it can perform well for multiple faces.Considering the proportion of different chroma of skin color, we proposed a method of skin detection based on improved YCbCr color space. Instead Cb of Cg. Experimental results prove that skin pixel have better clustering in improved YCbCr color space.Under simple skin underground, change axis on YUV color space, in order to increasing distribution deference between skin sample and lip sample, lastly, using the geometry feature of lip on human face to locate face. Experimental results prove that the method can provide lower false rate,and achieve good detection performance in multi-face image. Lower complexity suits to reality.更多还原