结合零空间法和F-LDA的人脸识别算法  

Face recognition combining the null space approach and the fractional LDA

作　　者：王增锋[1] 王汇源[1] 冷严[1] 

机构地区：[1]山东大学信息科学与工程学院,山东济南250100

出　　处：《计算机应用》2005年第11期2586-2588,共3页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60172022)

摘　　要：线性判别分析(LDA)是一种常用的线性特征提取方法。传统LDA应用于人脸识别时主要存在两个问题:1)小样本问题,即由于训练样本不足引起矩阵奇异;2)优化准则函数并不直接与识别率相关。提出了一种新的能同时解决以上两个问题的基于LDA的人脸识别算法。首先,通过重新定义样本的类内散布矩阵和类间散布矩阵,提出了一种新的零空间法。然后把这种新的零空间法与F-LDA(Fractional LDA)算法相结合,得到一种对人脸识别更有效的特征提取方法。实验结果表明,这种新算法具有较高的识别率。Linear discriminant analysis is one of the most popular linear projection techniques for feature extraction. The traditional LDA-based methods suffer from two main disadvantages： 1） the small sample size （SSS） problem; 2） their optimization criteria are not directly related to the classification accuracy. A new LDA-based technique was proposed, which could overcome the above shortcomings. First, by redefining the within class scatter matrix and the between class scatter matrix, a new null space method was presented. Then, combined the new null space method with F-LDA （fractional LDA） algorithm, a completely new feature extraction algorithm for face recognition was given. The experimental results show that the new LDA-based feature extraction algorithm improves the performance of face recognition significantly.

关 键 词：线性判别分析 零空间法 F—LDA(fractional—LDA) 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]