一种核化有全局性类内近邻保持正交算法  

A Kernel Within-class Neighborhood Preserving Orthogonal Algorithm with Global Structure

作　　者：林玉娥[1] 李敬兆[1] 梁兴柱[1] 林玉荣[2] 

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001 [2]哈尔滨工业大学航天学院,哈尔滨150001

出　　处：《光电工程》2013年第3期115-121,共7页Opto-Electronic Engineering

基　　金：国家自然科学基金(60975009;61170060);安徽省自然科学基金(1208085QF123;11040606M135);安徽省高等学校自然科学基金(KJ2012Z084;KJ2011A083);安徽理工大学中青年学术骨干资助项目

摘　　要：以主成份分析、局部保持及核映射理论为基础,提出了一种核化有全局性类内近邻保持正交算法。该算法的目标函数融合了样本的全局与局部信息,同时由于采用非线性映射及基向量的正交性限制,因此能够提取出更为有效的分类特征。但由于非线性函数的未知,因此无法直接对准则函数进行求解,对此根据核映射理论,本文将算法的不可解的准则函数转化为核空间上可解的准则函数,并给出了具体的理论推导及求解步骤。人脸库上的实验结果表明所提方法的有效性。Based on the principal component analysis, locality preserving projection and kernel mapping theory, kernel within-class neighborhood preserving orthogonal algorithm with global structure is proposed. The object function of proposed method takes the global and local information of the samples into account. At the same time, by using nonlinear mapping and the orthogonal restriction of base vectors, this algorithm can extract more effective classification features. Being unknown nonlinear function, the object function can not be directly calculated. But according to the kernel mapping theory, the equivalent object function of the proposed algorithm is established in the kernel feature space. The deducing process and solving steps are given. Experimental results on face database demonstrate the effectiveness of the proposed method.

关 键 词：核映射理论 类内近邻保持 目标函数 正交性 核空间 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]