摘要
针对核化图嵌入算法对于人脸识别等高维小样本问题存在着计算量大且所需存储空间多的缺点,提出了一种核化图嵌入算法的快速求解模型。该模型的思想是首先对原始样本进行降维处理,对此给出了定理1和2。两个定理证明了样本先进行降维处理的可行性,同时也表明这一过程是不损失任何有效鉴别信息的。然后再对新的低维样本按核化图嵌入算法进行计算。人脸库上的实验结果表明,所提模型不但减少了算法的计算时间,同时也保证了算法的分类识别率。
Kernel extension of graph embedding for the small sample size problem such as face recognition needs not only a lot of computation time but also very large memory cost,then this paper presented a fast model for kernel extension of graph embedding.Firstly,it reduced the original samples into a lower space,which was feasible according to theorem 1 and theorem 2.Two theorems also show that this dimension reduction process is not losing any discriminant information.Then it computed the new low dimensional samples by kernel extension of graph embedding.The numerical experiments on facial database show that the proposed model not only reduce the computational time but also ensure rate of recognition in classification.
出处
《计算机应用研究》
CSCD
北大核心
2012年第12期4758-4760,共3页
Application Research of Computers
基金
国家自然科学基金资助项目(60975009
61170060)
安徽省自然科学基金资助项目(1208085QF123
11040606M135)
安徽省高等学校自然科学基金资助项目(KJ2012Z084
KJ2011A083)
关键词
核化图嵌入算法
小样本问题
模型
鉴别信息
分类
kernel extension of graph embedding
small sample size problem
model
discriminant information
classification