摘要
为了将局部信息有效地运用到特征抽取并提高算法的鲁棒性,同时考虑到在人脸识别应用中出现的高维小样本问题,提出了一种基于局部均值的广义散度差无监督鉴别分析。该方法利用样本的非局部均值散度与倍的局部均值散度之差作为鉴别函数,不仅保留了样本分布的局部信息,而且避免了局部均值散度可能奇异的问题,并给出了算法的识别率随模型参数变化的曲线。YALE和FERET人脸数据库上的实验结果表明了该方法的有效性。
To effectively use the local information in feature extraction and improve the robustness, and simultaneously consider the high-dimensional small sample size problem in face recognition application, a method called generalized scatter difference unsupervised discriminant analysis based on the local mean is proposed. It utilizes the difference of between non-local mean scatter and C times local mean scatter as the discriminant function, so that the local information of sample distribution not only is preserved, but the problem that the local mean scatter may be singular is avoided. Besides, the recognition rate curves of the proposed method with the variation of model parameter C are illustrated. Experiments on YALE and FERET face image database validate its effectiveness.
出处
《计算机工程与设计》
CSCD
北大核心
2011年第7期2482-2484,2489,共4页
Computer Engineering and Design
基金
国家自然科学基金项目(60875004)
江苏省高校自然科学基金项目(07KJB520133)
江苏省自然科学基金项目(BK2009184)
关键词
特征抽取
局部均值
非局部均值
广义散度差
人脸识别
feature extraction
local mean
non-local mean
generalized scatter difference
face recognition