文档介绍：基于模糊线性判别分析的人脸识别算法设计
学院
专业
班级
学号
姓名
指导教师
负责教师
摘要
人脸识别技术是生物识别技术的一种,以其直接性,唯一性,方便性等特点,在公安,海关,交通,金融,视频会议,机器人的智能化研究等方面得到了越来越广泛的应用。人脸识别技术是模式识别领域中的一个前沿课题。在过去的几十年里,研究者尝试利用计算机来模仿人类识别人脸的能力,并提出了很多人脸识别的有效算法,利用不同技术提高了人脸识别算法的平均识别率。本文着重讨论一种把特征脸和模糊线性判别分析(FLDA)算法结合起来进行人脸识别的方法。
该方法利用主成分分析(PCA)方法求得训练样本的特征空间,然后在此基础上计算FLDA算法的特征子空间,进一步对特征脸空间降维。经过FLDA降维后的子空间中,同一类别的样本尽可能靠近,不同类别的样本尽可能分散(即降维后同一个人的人脸图像尽可能的靠近,不同人的人脸图像尽可能的分散开)。模糊LDA方法引入了模糊技术来优化特征提取,利用隶属度信息来描述样本的分布信息,能得到一个更好的类中心位置估计。应用于Yale及ORL人脸库的实验结果表明,该算法具有较高的识别率。
关键词:人脸识别;主成分分析;模糊线性判别分析;特征脸
Face Recognition Algorithm based on Fuzzy linear discriminant analysis
Abstract
Face recognition technology is a kind of Biological recognition technology. With its immediacy, uniqueness and convenience, etc. It gets more and more widely used in terms of public security, customhouse, traffic, finance, video conference, the study on robot’s intelligence. Face recognition technology is a frontier topic in the field of pattern recognition. In the past few decades, the researchers tried to use puter to imitate human's ability to recognize faces, and a lot of effective algorithm of face recognition was proposed, and they used different technology increased the average recognition rate of face recognition algorithm. This paper focuses on a face recognition method bining with the ponent analysis and fuzzy linear discriminant analysis (FLDA) algorithm.
This method obtains the characteristics space of the training sample with the ponent analysis(PCA) algorithm, then on the basis of this calculation, get another FLDA’S feature subspace which has lower dimensions. In this FLDA’s feature subspace, samples of the same category are as near as possible, different types of sample are as disperse as possible (In other words, after the dimension reduction, the same person face image are as near as possible, the different human face image are as far as possible). The fuzzy technology is used in fuzzy LDA to optimize feature extraction, it can get a better clas