【作者】 时卫霞；

【导师】 戴青云；

【作者基本信息】 广东工业大学 ， 信号与信息处理， 2008， 硕士

【摘要】 生物识别是一种根据人体自身的生理特征(如指纹、脸像、虹膜等)和行为特征(如笔迹、声音、步态等)来识别身份的技术。由于人脸作为一种高普遍性的、可非接触式采集的重要生物特征,可以应用在多种不同的安全领域。特别是很多场合如银行,图书馆,酒店等都需要入口控制,防止非法进入,以往都采用人工核查身份证件等方法,费时费力,随着平安城市建设的蓬勃开展,越来越多的摄像头被运用到治安中,这将使得人脸的采集极其方便,自然地,对人脸识别设备的需要也越来越强烈。本文将数字图象处理结合模式识别的原理和方法应用到入口管理领域,建立了一套适用于金融营业网点的基于人脸识别的视频安保系统。实现了金融营业网点门卫系统的自动化、高效化。视频中人脸识别可以划分为三个环节:单帧人脸检测、视频中人脸跟踪,对人脸图像序列的识别。本文分别对这三个环节展开研究,并建立了一套适用于实际金融营业网点的视频安保系统的人脸检测、识别算法系统原型。根据其入口控制的实际工作环境,对于人脸的检测,本文提出了基于肤色信息和人脸特征做验证的人脸检测方法,即由背景更新的差分检测出运动目标,然后从定位的运动子图象中利用肤色分割定位人脸,肤色分割结合人眼来验证人脸。对于人脸的跟踪,本文提出了基于运动物体质心和滤波器预测的跟踪方法,利用运动目标的质心代替运动物体和卡尔曼滤波器来预测物体的运动方向和速度,以此来达到跟踪物体的目的。在人脸识别部分,提出了将下颌作为人脸识别的新特征与其他几何特征相结合的人脸识别方法。通过精确定位人脸中眼球及眼角、嘴巴的位置与人脸的轮廓线相结合作为提取的特征构造特征矢量,最后利用所设计的最近类中心分类器进行识别。实验结果表明,本文所采用的算法简单并且在一定的限制条件下,系统取得了令人满意的效果。更多还原

【Abstract】 Biometrics means automatically identification of an individual based on their biological features such as fingerprint, face and iris or behavioral traits such as signature, speech and gait. Facial images are probably the most common and on-intrusive biometric characteristic for personal identification. Face recognition can be applied to the entrance control in various fields, such as banks, libraries, etc. In this paper,to makes the entrance control more automatic and more effective,a real-time video security system based on face recognition is proposed using methods of image processing and the pattern recognition.Video face recognition consists of three parts: human face detecting in single frame, human face tracing of video, identification of human face image sequences. This dissertation is focused on these three parts and practical video security system is set up for financial business outlets management. A new method of face detection applied to the system is proposed. Considering the situation of the system, difference detection and skin color mapping are used to detect and locate the faces .A method is proposed to solve target tracking problem of the moving object under occlusion efficiently that is a combination of the center of mass and Kalman filter. The simulation experiments show that the tracking is robust to partial or short-time occlusion. As to the face recognition, the method is based on geometrical features in part by a chin contours as new features. In the process of feature points extraction, variance projection and edge detection are used by precise positioning of the eyes , mouth, nose and chin contours of face. At last, similarity of two face image feature vectors is calculated, and a proper threshold is selected to judge whether the two face images belong to the same person.The experiments proved that this method has simpler arithmetic and higher recognition rate.更多还原