导　　师： 樊春玲

学科专业： H1101

授予学位： 硕士

作　　者： ;

机构地区： 青岛科技大学

摘　　要： 随着社会的发展，各方面对快速有效的自动身份验证的要求日益迫切。由于生物特征是人的内在属性，具有很强的自身稳定性，因此是身份验证的最理想依据。人脸作为图像与视频中最重要的视觉对象之一，在计算机视觉、多媒体技术、模式识别等研究中占有重要的地位。 本课题是在青岛科技大学虚拟仪器实验室的现有设备的基础上，研究人脸检测算法，人脸识别算法，设计并实现了面向视频流的教室场景的人脸识别系统。该系统能够有效的提高远程教学环境中，教师与学生的互动交流。 论文采用了机器视觉技术与虚拟仪器相结合的方式，充分利用LabVIEW图形化编程，人机界面美观，对图像处理易于操作的优点，同时调用VC++6.0生成的动态链接库完成系统关键算法的设计，提高了系统运行的速度。图像采集使用NI公司的摄像头和采集卡，接口简单，传输速率高，实时性强。 人脸检测部分运用肤色和模板匹配相结合的算法，首先在YCbCr色彩空间下，利用肤色信息将彩色图像中的肤色区域和非肤色区域分开，通过计算得到表示肤色相似度的相似度图，然后对此图像阈值分割，进过形态学处理，候选区域筛选，模板匹配，最终确定图像中人脸的位置。 人脸识别部分，首先图像经过人眼检测，人脸正规化，人脸掩模，滤波等预处理后，提取图像LBP特征，然后根据最短距离的K近邻法则对待识别样本分类。 With the development of modern society, requirement of the fast and efficientautomatic authentication is increasingly urgent. Biological characteristics are intrinsicproperties along with its’ strong stability, hence they can be the best basis ofauthentication. Face, which acts as one of the most important visual object in imagesand videos, occupies an important position in computer vision, multimedia technology,and pattern recognition research. This research is on the basis of the existing equipment-Virtual InstrumentLaboratory of the Qingdao University of Science and Technology, including the studyof face detection algorithm, face recognition algorithm, the design and realization ofthe face recognition system for the video stream used to classroom scene.The systemcan effectively improve the demote learning environment, which can contribute to theinteraction between the teachers and students. This paper uses a combination of machine vision technology and virtualinstruments, it can make full use of the metrics of LabVIEW such as graphicalprogramming, the friendly man-machine interface, easy operation for imageprocessing, and dynamic link library of VC++6.0is called to complete the design ofthe critical part of the algorithms to improve the running speed of the system. WithNI’s camera, Image Acquisition has the following advantages, interface simple, hightransfer rate, and better real time quality. Face detection use a combination of color and template matching algorithm,firstly in the YCbCr color space, with complexion, images are separated into skinregions and non-skin regions, then calculate the color similarity chromaticity diagram,and then split the resulting chromaticity diagram into the grayscale image, resulting inthe candidate region of the face detection, and Euler number calculation in the regionto narrow the search range, and then use template matching, and ultimately to identifythe location of human faces in the images. Face recognition, after the preprocessingsuch as human eye detection, face regularized, face mask, filtering, we can extract LBPfeature, and then use the nearest neighbor rule to identify classify faces.

关 键 词： 人脸检测 人脸识别 特征抽取 动态链接库

分 类 号： [TP]

领　　域： [自动化与计算机技术]