【作者】 耿续涛；

【导师】 刘辉；

【作者基本信息】 昆明理工大学 ， 计算机软件与理论， 2007， 硕士

【摘要】 人脸检测及识别技术是机器视觉与模式识别领域最有挑战性的研究课题之一，它是一个涉及到模式识别、计算机视觉，自然语言理解、图像处理等多门学科的综合课题，随着社会的发展、科技的进步，对方便、可靠、自动身份鉴别的实际需求日益迫切，人脸识别技术也是目前研究较多的热点问题之一。本文对人脸检测及识别技术进行了研究，并进行了系统原型的设计，主要的工作如下：(1)详细研究了Adaboost快速人脸检测的方法，分析了该方法所使用的矩形特征及积分图的计算方法，进而引入了Rainer Lienhart等人提出的扩展的矩形特征，利用这些矩形特征构造了若干个弱分类器，接着把这些弱分类器级联成为一个强分类器，设计了人脸检测子系统。通过摄像头对该子系统进行测试，可以看出该方法检测速度很快。在检测的同时，对检测的人脸进行灰度变化、直方图均衡化等处理，使得图片更适合于人脸识别。(2)分析了人脸识别中比较经典的方法——基于主成份分析的特征脸方法。通过对图像进行变换得到特征脸，进而可以把所有的图片都由这数个特征脸图片的线性组合表示。并且详细分析了用于人脸分类的方法，以及使用的度量函数，并介绍了在本文中所使用的方法。基于上述方法，本文实现了人脸识别的子系统，并对ORL库进行了实验。(3)把人脸检测及识别两个部分有机的组合在一起，构建了一个自动完成人脸检测及识别的较为完整的系统原型。把该系统用于实际的测试中，每秒钟可以进行5帧图片的检测及识别，然后对这5个结果进行统计，得出最后的识别结果，在实验中取得了较快的速度和较好的识别率。更多还原

【Abstract】 Face detection and recognition technique is one of the most challenging subjects in the fields of machine vision and pattern recognition. It is a comprehensive topic which involves in the pattern recognition, the computer vision, the understanding of natural language, image processing and so on. Along with the development of society and the advancement of technology, the actual needs for convenient, reliable and automatic identification are increasingly urgent, and face recognition technology is also a hot topic in research currently.In this thesis face detection and techniques are studied, and a system prototype is designed. The main work is as follows:(1)Adaboost algorithm which is a rapid face detection method is studied in detail, the rectangular features and the method of calculating Integral Image are analyzed. And then the expansion of the rectangular features proposed by Rainer Lienhart is added. Some weak classifiers are constructed using these rectangular features, and then a strong classifier is constructed by cascading the weak classifiers. Using the classifier, a subsystem of face detection is designed. Through the test using a camera, it can be seen that this system is very quickly. At the same time, the face area is processed through gray adjust and histogram equalization, to make the face area is more suitable suited for face recognition.(2)Eigenface based on Principal Component Analysis is analyzed, which is a classic algorithm for face recognition. Eigenfaces are achieved by transforming the images, and all the face pictures can be expressed by linear combination of these several eigenfaces. And the classification methods of human faces and specific measuring function are analyzed in detail. Based on the above, a face recognition subsystem is realized inthis thesis and a experiment was carried on the ORL.(3)The face detection subsystem and recognition subsystem are combined together, to form an automatic face detection and recognition system which is a relatively complete prototype. Under actual test, this system can carry out 5 pictures of detection and the recognition, and then the 5 results are carried on statistical analysis, to obtain the final recognition result. And it has achieved quick speed and better recognition rate.更多还原