基于非负矩阵分解的人脸识别研究

发布时间：2018-09-09 12:51

【摘要】：人脸识别是在图像或视频流中进行人脸的检测定位和识别。近年来由于计算机运算速度的飞速发展使得图像处理技术有了更好的硬件支持,进而人脸识别技术也成为计算机视觉与模式识别领域中非常活跃的研究课题,在身份验证、可视通信、智能监控、公安档案管理等很多方面都有着广泛的应用。广义上的人脸识别过程包括人脸的检测定位和人脸的匹配识别两个部分。本文先综述了人脸识别的研究背景和研究现状,并详细的通过分类介绍了人脸检测和人脸识别中的主要方法。然后在深入的研究了已有算法的基础上,对算法进行了调整和融合,并完成了一个人脸识别软件的设计和实现。在进行人脸检测之前,本文先探讨了人脸图像的预处理,主要介绍了人脸图像的灰度变换和直方图均衡化。在介绍原理的基础上,实验选取了一幅静态人脸图像对操作效果进行了直观描述。人脸检测环节,本文采用基于Adaboost算法的人脸检测。Adaboost算法是采用分类器的思想,把人脸检测问题看成一个只有两个类的分类问题,即将图像分为“人脸”和“非人脸”两大类。通过对人脸样本图像进行训练得到的级联分类器来对人脸进行检测,在此基础上实现了基于Adaboost算法的人脸检测软件。选取了ORL人脸数据库和MIT人脸数据库中的图像样本进行了统计实验,观察了级联分类器个数对检测率的影响。人脸识别部分,先介绍了NMF算法的原理,之后又介绍了几种基于NMF的改进算法。最后在分析了非负矩阵分解和其改进算法的基础上,采用了稀疏约束的非负矩阵分解算法进行特征提取,通过对基矩阵稀疏度的改变探讨了其对人脸识别率的影响,并在ORL人脸库上进行了实验。本文的人脸识别和检测部分的实验以及最后给出的软件都是采用Windows平台,使用Visual C++6.0结合OpenCV实现的,实验结果表明本文采用的算法在检测率和识别率上都满足要求。
[Abstract]:Face recognition is the detection, localization and recognition of face in image or video stream. In recent years, because of the rapid development of computer computing speed, image processing technology has better hardware support, and face recognition technology has become a very active research topic in the field of computer vision and pattern recognition. Visual communication, intelligent monitoring, public security file management and many other aspects of a wide range of applications. The process of face recognition in a broad sense includes two parts: face detection and location and face matching recognition. In this paper, the research background and current situation of face recognition are reviewed, and the main methods of face detection and recognition are introduced in detail by classification. Then, on the basis of deeply studying the existing algorithms, the algorithm is adjusted and fused, and a face recognition software is designed and implemented. Before face detection, this paper discusses the preprocessing of face image, mainly introduces the gray level transformation and histogram equalization of face image. On the basis of introducing the principle, a static face image is selected to describe the operation effect directly. In the aspect of face detection, this paper adopts the idea of classifier for face detection based on Adaboost algorithm, and regards the problem of face detection as a classification problem with only two classes, that is, the image can be divided into two categories: "face" and "non-face". The face detection software based on Adaboost algorithm is implemented based on the cascade classifier which is trained to the face sample image. The image samples from ORL face database and MIT face database are selected for statistical experiments, and the effect of the number of cascaded classifiers on the detection rate is observed. In the part of face recognition, the principle of NMF algorithm is introduced, and then several improved algorithms based on NMF are introduced. Finally, based on the analysis of the non-negative matrix decomposition and its improved algorithm, the sparse constraint non-negative matrix decomposition algorithm is used to extract the features, and the effect on the face recognition rate is discussed by changing the sparse degree of the base matrix. Experiments are carried out on ORL face database. The experiments of face recognition and detection in this paper and the software presented in the end are implemented on Windows platform and Visual C 6.0 combined with OpenCV. The experimental results show that the detection rate and recognition rate of the algorithm in this paper meet the requirements.
【学位授予单位】：西北师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41

【相似文献】