摘要人脸检测在开始的时候被用来支持人脸识别[1]，检测到人脸后对人脸区域识别。但是由于近几年各种相关应用的发展使得人脸检测成为单独的研究课题，受到社会和研究人员的重视。目前为止，人脸检测应用的范围已经超过人脸识别的范围，人它已经广泛应用于身份的验证、人脸识别、表情识别和视频监控等领域。63861

本文首先简介了一些常用的人脸检测算法[9]，然后详细介绍基于haar特征值的Adaboost方法，简述特征值，积分图，级联等相关知识。依此算法训练出来的级联强分类器在人脸检测时准确度和速度都有比较好表现[18]，但是在训练时需要较长的时间，而且它对光线的方向，强弱比较敏感，外部环境比较差的情况下检测效果会有较大的下降。详细介绍级联分类器的训练过程，并且提出了不少训练时遇到的问题及解决方法，最后实现一个用于实时人脸检测与人眼定位的程序，可以用它来实现实时系统中的人脸检测与人眼定位，以及快速寻找视频中出现的人脸。

毕业论文关键词： 人脸检测，haar分类器，Adaboost，分类器训练，人眼定位

Abstract Face detection in the beginning was used to support face recognition, which recognizing the face after detecting it.However, with a great change of related applications in recent years, face detection constitute a separate research topic.So far,causing much attention of the community and researchers.it has been widely used in human identity verification, face recognition, face recognition and video surveillance.

　　Firstly, we introduce some common face detection algithms some details based on haar eigenvalues’Adaboost method, then outline eigenvalues, integral image, Cascade other related knowledge.Based on this algorithm trained cascaded strong classifier, face detection has good performance in accuracy and speed.However, the training takes a long time, and it depends on direction of the light, the strength of sensitive and when external environment is poor ,it will have a great drop on detection 

Results. This thesis detail cascade classifier training process, and put forward a number of training problems and solutions, and finally to achieve a real-time face detection for the human eye positioning program, it can be used to achieve real-time systems face detection and the human eye positioning, and quickly find the human face appearing in the video.

Keywords: Face Detection, Haar classifier, Adaboost, Classifier training, Eyes Location

目录

摘要 1

Abstract 2

目录 3

1  绪论 4

1.1 研究背景和意义 4

1.2 研究现状 5

2  常见人脸检测算法 6

2.1 概述 6

2.2 肤色区域分割与人脸验证方法[7] 6

2.3 基干启发式模型的方法 6

2.4 基于特征空间的方法[27] 6

2.5 基于支持向量机的方法[32][33][34] 6

2.6 基于人工神经网的方法[10] 7

 3  Haar分类器 7

3.1 Haar-like特征 7

3.2 积分图方法 8

3.3 Adaboost算法 9

3.3.1算法概述AdaBoost[14] 9

3.3.2弱分类器