基于PCA和BP神经网络的人脸识别方法研究

发布时间：2018-01-15 11:49

本文关键词：基于PCA和BP神经网络的人脸识别方法研究　出处：《哈尔滨理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着科学技术和互联网的发展和应用,平安城市和智慧城市的概念深入人心,各种各样先进的高清摄像头逐渐应用在城市的监控设备系统中,监控系统迫切的需要安装使用效率高、检测识别耗时少和交互体验良好的人脸识别系统来进行智能监控。人脸这一人类固有的生物特征具备诸多如采集方便、不变性和不会被盗取的优点,是人脸识别系统成功应用在各个领域的关键所在。如何改进人脸识别技术算法来提高检测识别速率、效率以及系统的操作易用性来满足城市的监控系统的需求是本文研究关注的重点。原始图像的采集和检测、人脸的预处理、特征的提取和匹配分类是人脸识别系统当中四个缺一不可的关键步骤,本文集中对这四个步骤进行研究分析。首先,使用基于Haar-like特征(Haar-based)的人脸检测分类器检测处理已采集的原始图像,对检测出人脸的图像进行裁剪处理。其次,预处理裁剪出的人脸图像,详细步骤有:灰度化处理、去噪声处理、校正处理和归一化处理。在特征提取阶段,在使用LBP、PCA、A-PCA这三种特征提取算法在人脸库上做了大量对比实验及详细分析后可得,本文提出的基于类别模式的PCA算法(A-PCA)具有较高识别率和鲁棒性。在匹配识别阶段,BP神经网络具有很强的学习能力,可以处理非线性的、含有噪声的信号,适用于处理非线性分类识别,结合A-PCA算法在Yale、ORL和AR人脸数据库和自建人脸库上进行实验表明:A-PCA+BP神经网络具有较高的识别率,对光照变化,表情变化有较高的鲁棒性。最后,本文使用MATLAB软件和OpenCV开源库搭建实现A-PCA+BP神经网络的人脸识别系统,系统能够实时检测人脸与识别,改进的算法得到了实现,提高了识别率,具有良好的交互体验,在平安智慧城市的监控系统中具有一定的实用价值。
[Abstract]:With the development and application of science and technology and the Internet, the concept of "Ping an City" and "Smart City" is deeply rooted in the people's mind, and a variety of advanced high-definition cameras are gradually applied in the surveillance equipment system of the city. The monitoring system urgently needs to install the face recognition system with high efficiency, less detection time and good interactive experience to carry out intelligent monitoring. Face, which is the inherent biological feature of human, has many advantages such as easy to collect. The advantages of invariance and non-theft are the key to the successful application of face recognition system in various fields. How to improve the face recognition algorithm to improve the speed of detection and recognition. The efficiency and ease of operation of the system to meet the needs of the urban monitoring system is the focus of this study. The acquisition and detection of original images, face pretreatment. Feature extraction and matching and classification are the four indispensable key steps in face recognition system. This paper focuses on these four steps. First of all. The face detection classifier based on Haar-like feature is used to detect and process the original image, and the image of the detected face is clipped. Secondly. The detailed steps are grayscale processing, noise removal, correction and normalization. In the feature extraction stage, LBPPCA is used. A-PCA three feature extraction algorithms in the face database of a large number of comparative experiments and detailed analysis can be obtained. The class pattern based PCA algorithm proposed in this paper has high recognition rate and robustness. In the matching recognition stage, BP neural network has a strong learning ability and can deal with nonlinear. The signal with noise is suitable for processing nonlinear classification and recognition, combined with A-PCA algorithm in Yale. The experiments on ORL and AR face database and self-built face database show that the BP neural network has high recognition rate and robustness to illumination change and expression change. Finally. This paper uses MATLAB software and OpenCV open source library to build a face recognition system based on A-PCA BP neural network. The system can detect and recognize face in real time. The improved algorithm has been implemented, improved the recognition rate, and has a good interactive experience. It has certain practical value in the monitoring system of the peaceful and intelligent city.
【学位授予单位】：哈尔滨理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP183

【相似文献】