【作者】 丁宏伟；

【导师】 赵宏伟；

【作者基本信息】 吉林大学 ， 计算机技术（专业学位）， 2020， 硕士

【摘要】 随着社会的不断发展和科学技术的不断进步,当今世界越来越需要一种能快速确定人们身份信息的身份验证方法,人脸识别以其采集信息方便快速、非侵入性等等优点,使其在我们身边应用越来越普遍,如今已经成为科研的最热门研究领域之一,越来越多的研究者投入到人脸识别算法研究系统开发的研究中去。随着越来越深入的人脸识别算法的研究,人脸识别系统的性能也越来越好,应用也越来越广泛。不同的应用环境对系统有着不同的功能要求,本文基于opencv开发了一个可以对摄像头拍摄视频中的人进行实时的人脸检测和人脸识别的人脸识别系统,系统的主要界面有三个即主界面、人脸录入界面、人脸识别界面,系统的操作步骤简单、明确,在兼顾准确性的前提下实现了实时性。人脸检测模块中使用了opencv中提供的基于Haar特征的Adaboost人脸检测器,针对该人脸检测器对人脸的误检这一现象提出了肤色人眼二次验证的改进人脸检测器,改进后明显降低了人脸检测器的误检率,且对有遮挡人脸检测有了一定的提升。另外通过对实验室同学人脸图像的采集和对寝室以及实验室环境的采集结合已有的人脸数据库建立了自己的人脸数据库和背景环境数据集,在此基础上自行训练了人脸检测器。人脸图像经过人脸检测器从帧图像上截取下来之后,需要进行图像的预处理,增强人脸图像中的人脸特征,降低环境因素对人脸识别准确度的影响,预处理之后将图像存储在与其标签命名相同的文件夹中,以备人脸识别模块中分类器的训练。人脸识别模块分为特征提取部分和人脸匹配部分,特征提取部分详细的介绍了局部二值模式(LBP)特征提取算法,针对LBP算法对图像特征利用不足的缺点提出了双重LBP算法,并通过实验证明了在人脸样本较少的情况下相较于传统的LBP算法改进算法提升了人脸识别的准确性,通过实验证明了改进算法提升了对偏转角度人脸识别的准确率。人脸匹配部分使用的是在小样本上分类表现良好的SVM(支持向量机)算法,将双重LBP特征提取算法与SVM分类算法结合在一起实现了人脸识别模块。更多还原

【Abstract】 With the continuous development of society and the progress of science and technology,the world is more and more people need a can quickly determine the identity authentication,face recognition for its convenient rapid,noninvasive,and so on gathering information advantage,make its application more and more common around us,now has become a scientific research one of the most popular research fields,more and more researchers into the face recognition algorithm in the study of system development.With more and more research on face recognition algorithms,the performance of face recognition system is getting better and better,and the application is becoming more and more extensive.Different to the functional requirement of the system has a different application environment,this article is based on opencv developed a camera can be the man in the video real-time face detection and face recognition of face recognition systems,there are three main interface system main interface,face input interface and face recognition interface,the operation of the system steps is simple,clear,in both the accuracy of the premise to realize the real time.Provided during the face detection module is used in the opencv,Adaboost face detector based on Haar feature of face for the face detector mistakenly identified the phenomenon of color eye secondary validation of improved face detector,the improved significantly reduces the error detection rate of face detector,and the covered face detection has a certain improvement.In addition,by collecting face images of students in the laboratory,and collecting dormitory and laboratory environment,I established my own face database and background environment data set based on the existing face database,and trained face detector on this basis.Face image after face detector intercept down from the frame,the need for image preprocessing,the facial features in a face image,and reduce the effects of environmental factors on the accuracy of face recognition,after preprocessing the image stored in the same folder name,its tag has been training classifier in face recognition module.Face recognition module is divided into feature extraction part and face matching part,feature extraction part detailed introduces the local binary pattern(LBP)feature extraction algorithm,aimed at the shortcoming of lack of LBP algorithm for image features using dual LBP algorithm was presented,and proved through the experiment in the face of samples under the condition of less than traditional LBP algorithm improved algorithm improved the accuracy of face recognition,through the experiment proves that the improved algorithm increases the accuracy of deflection Angle face recognition.The face matching part uses SVM(support vector machine)algorithm with good classification performance on small samples.The dual LBP feature extraction algorithm and SVM classification algorithm are combined to realize face recognition module.更多还原