摘 要

身份识别的安全性和智能化一直是备受关注，特别是如今互联网、通信等科技迅速发展，尤为重视它的研究，其中人脸识别系统因为具有简便性、唯一性、应用范围广等优点脱颖而出，另外由于图像处理技术和机器学习算法都有进一步的研究成果，在理论上给予了足够的技术支持，而人脸识别系统相对其他识别系统来说不需要太过复杂的硬件和软件来实现识别，因此，人脸识别具有很好的应用前景。

本设计的视频人脸识别系统主要有两个模块，分别是人脸检测和人脸识别模块，其对应的是人脸检测和人脸识别技术。人脸检测是指通过摄像头摄取视频中的图像，使用相关的人脸检测技术，检测出目标人脸，并将其发送到人脸识别模块中，发送的信息包含人脸的位置和大小等信息；人脸识别是指对人脸检测模块检测出的人脸信息根据相应的人脸识别方法进行识别和判断，利用预存的人脸信息库判别其检测目标是否为预先设定的人脸，从而得出其关联的人物信息，另外不同的识别技术判断的根据和出来的效果也会不同。除了以上两个最主要模块以外，还需要注意图像的预处理，由于实际进行识别时会有一些难以避免的干扰，如光照、角度、旋转等因素，进行图像的预处理尽可能的排除干扰因素的影响很有必要。

1.对人脸识别技术的目的、意义和研究现状进行了研究和阐述，同时对人脸识别有一个总体上的认识，包括人脸识别的过程和它的优缺点以及它所需要的技术等内容。

2.对人脸识别每一过程总体上分类概述，每一类可以使用的技术也作一较为详细的概述。

3.对本设计做一个总体设计，选择一个较为准确而有效的方案，并将方案中需要使用的技术做一个清晰而明确的表达，同时将方案中实现的过程有一个较为准确的了解。

4.对总体设计上所用到的技术做一个详细的阐述。

图像的预处理模块，一般有消除噪声、灰度归一化、几何校正、滤波变换、光照补偿等方法，根据实际情况采取不同方法，适当的组合效果有可能会更好，不过也不能都用上，这会有反效果，因为各种方法之间会有影响；

人脸检测模块，采取Adaboost算法，将算法原理向读者阐明，简单地说就是采用 adaboost 算法训练出的人脸最优分类器对待测图像进行人脸检测，其标准形式作为首选方案；

人脸识别模块，将检测到的信息在进行特征提取，采取PCA（主成分分析）算法进行特征提取，PCA有降维的作用；采用基于支持向量机（SVM）的特征分类识别算法进行分类识别，对不同情况采用不同的SVM方法，这一阶段将提取到的矩阵信息与在之前已经存入的人脸库中的信息进行人脸比对和匹配，达到识别的目的，从而判断和得出被检测者的身份。

本设计采用的视频人脸识别系统，采用外设摄像头摄取视频中的短时间内的5帧来作为被检测者的图像，采用ORL人脸库和视频所截取的被检测者图像作为检测图像，使用opencv和MICROSOFT VISUAL STUDIO 2010软件来编程和实现。结果表明：本设计报告采用的算法所实现的人脸识别系统拥有着较高的识别率和检测速度。

关键词：人脸识别；图像的预处理；Adaboost；PCA；SVM

Abstract

Identification of the security and intelligence has been concern, especially in today's Internet, the rapid development of communication technology, particularly pays attention to the research, which face recognition system because is simple, uniqueness and application scope wide advantages come to the fore, also due to the image processing technology and machine learning algorithms have further research, in theory gave enough technical support, and face recognition system relative other recognition systems don't need too complex hardware and software to achieve recognition. Therefore, face recognition has a good application prospect.

The design of the video face recognition system has two main modules, namely face detection and face recognition module, which corresponds to the face detection and face recognition. Face detection is an image through the camera in the video, using face detection technology, detect the target face, and send it to face recognition module, send information includes the size and location of the face information; face recognition is the face information detected by the face detection module to identify and judge according to the corresponding methods of face recognition, face information database using a pre stored target detection is to judge its face set in advance, so as to obtain the related information of characters in addition, different recognition according to the judgment and the result is also different. In addition to the above two main modules, also need to pay attention to image preprocessing, by There are some difficult to avoid interference, such as light, angle, rotation and other factors in the actual identification. It is necessary to eliminate the interference factors as far as possible.

1. The face recognition technology of purpose, significance and research status of research and the elaboration, and it also has a general understanding of face recognition, including face recognition process and its advantages and disadvantages and the technology, and so on.

2.overview of the overall classification of human face recognition, each class can be used to make a more detailed overview of the technology.

3. the design do a overall design, a more accurate and effective scheme selection and scheme requires the use of technology to do a clear and explicit expression, also will have a more accurate understanding of the scheme is implemented in the process.

4.on the overall design of the technology used to do a detailed exposition.

Image preprocessing module, generally have to eliminate noise, gray normalization, geometric correction, transform filter, light illumination compensation method, according to the actual situation take different methods, appropriate combination effect can be better, but also can not use, which will have the opposite effect, because for between the various methods will be affected;

Face detection module, take AdaBoost algorithm, will algorithm principle to the reader elucidated, simple to say is to use AdaBoost algorithm to train the face optimal classifier treatment image for face detection, the standard form as a preferred solution;

Face recognition module, the detected information in feature extraction, PCA (principal component analysis) algorithm for feature extraction, PCA to reduce the dimensionality of the role; used for classification and recognition based on the support vector machine (SVM) classification feature recognition algorithm, for the SVM method is not the same, this stage to extract the information matrix and before has been deposited in the face database information were face alignment and matching, to identify the purpose to judge and get detected by identity.

The design of the video based face recognition system, the peripheral camera to capture a video in a short time of 5 frames as detected by image, using the ORL face database and video to the interception of the detected image as image detection, using OpenCV and Microsoft Visual Studio 2010 software programming and implementation. The results show that the design report using the algorithm to achieve the face recognition system with high recognition rate and the speed of detection.

Key Words：Face recognition; image preprocessing; PCA; Adaboost; SVM