在飞速发展的当代社会中，安全防范问题越来越突出，海关、机场、车站等人员频繁流动场地对实时监控识别系统的性能需求大大提高，因此，开展基于视频流的人脸检测识别研究将有着特殊的应用价值和广阔的发展前景。 人脸检测识别技术是一种典型的生物识别技术，涉及人工智能、图像处理等诸多科学理论，拥有广泛的科研价值。本文分别从视频图像处理、人脸检测和特征提取及识别三个阶段对其做了深入研究，实现了动态视频图像的人脸检测及识别。 视频图像处理阶段，采用灰度归一化，对视频RGB图像进行灰度化处理；通过光照归一化，削弱光线对图像的不利影响；通过尺度归一化，放缩用于检测人脸的视频图像以及特征提取的人脸图片；通过滤波，消除人脸图片的噪声信息。在人脸检测阶段，分析研究了AdaBoost检测算法，其中包括Haar特征提取、分类器训练及级联等，而且合成了人脸分类器，并且优化了算法的检测速率和检测率。在特征提取及识别阶段，重点分析了LBP算法，针对传统的LBP算子提取的脸部纹理特征不够完整的缺点，提出了双编码局部二值模式（Double Coding Local Binary Pattern，d-LBP）人脸识别算法；改进的算法在传统LBP算法的基础上，充分考虑了局部邻域内各个像素点灰度值之间的均值和幅值关系，从而较完整的提取人脸纹理特征；然后结合PCA维数约简分块提取人脸纹理特征并统计直方图，根据最近邻原则进行识别，并在ORL人脸库进行对比实验，实验数据表明，d-LBP算法拥有较强的纹理描述能力，能有效提高识别率。 基于视频流的人脸检测识别系统分为识别与注册两大模块，在Qt开发环境下设计实现了该系统；系统测试表明，本文所实现的系统对姿态、表情、光照及距离有一定的鲁棒性，拥有较高的识别率。

In the modern society of rapid development, security issues have become more and more prominent, and it is necessary for customs, airports and railway stations to improve the performance of real-time monitoring identification system because of the frequent movement of people. Therefore, study on the face detection and recognition system which is based on video identification will have a special value of application and a broad prospect of development. Face detection and recognition technology is a typical biological recognition technology, it relates to a lot of scientific theories, such as artificial intelligence, image processing and so on, and it has broad research value. This paper research face detection and recognition technology deeply from three respects, which are video image processing, face detection, face feature extraction and recognition, and realize the face detection and recognition based on the dynamic video image. In the video image processing stage, the color image is converted into a gray image by using the gray scale normalization; by the method of illumination normalization, it can weak the adverse impact on the image of the light; the video images which are used to detect human face can be zoomed in and out by the method of scale normalization, so are the face images; in order to eliminate the noise of face images, the method of noise filter can be used. In the Face detection stage, the AdaBoost face detection algorithm has been researched in detail, including the Haar feature extraction, classifier training and classifier cascade; and then the face classifier has been created with a lot of face and non-face sample, and also improves the rate of face detection. In the feature extraction stage, this paper makes a special effort on study of the LBP algorithm and proposes a new Double Coding Local Binary Pattern algorithm (d-LBP) to improve the weakness of traditional LBP algorithm, such as incomplete features extraction. The improved algorithm can completely and fast extract LBP texture feature, which succeed in taking full consideration of the relationship of amplitude and mean among pixel gray values. Then, the paper uses the d-LBP algorithm combined with PCA (Principal Component Analysis) to extract statistical characteristics in each small block of the original face image and get the vector of texture feature, and it fulfills the face recognition by using K Nearest Neighbor algorithm. Finally, experiments have been done in the face database of ORL, experimental results shown that the d-LBP algorithm has a good ability of identification and improves the recognition rate in some degree. The Face detection and recognition system based on video has been divided into face recognition module and face registered module. It has been designed and implemented in the development environment of Qt, and the experiment shows that the system has a good ability of identification, and a certain degree of robustness to posture, expression, illumination and distance.