摘要
"维度灾难"是基于内容的图像检索(CBIR,content-based image retrieval)技术需要重点解决的关键问题。局保投影(LPP,locality preserving projections)流形学习算法可以最大限度地保留图像的局部非线性结构,从而更能够保留图像的本质特征。利用LPP流形学习算法的特性,在CBIR框架下,提出了一种图像检索特征降维方法。实验结果表明,方法在保持与原始特征基本相当的检索性能情况下,特征比对的计算复杂度可以降低66.51%。
＂Curse of dimensionality＂ is a key issue needed to be solved in CBIR. The local nonlinear structure of image is retained in maximum by LPP manifold learning algorithm, and the essential characteristic of the im- age is kept well. According to the characteristic of LPP algorithm, an image retrieval feature dimensionality re- duction method based on manifold learning is proposed on the basis of CBIR. Experimental results demonstrate that the proposed method can achieve comparable retrieval performance compared with original features, while computational complexity of feature match is reduced by 66.51%.
出处
《测控技术》
CSCD
北大核心
2014年第12期8-10,15,共4页
Measurement & Control Technology
基金
国家自然科学基金项目(61372149
61370189
61100212)
教育部"新世纪优秀人才支持计划"项目(NCET11-0892)
高等学校博士学科点专项科研基金资助课题(20121103110017)
北京市自然科学基金(4142009)
北京市属高等学校高层次人才引进与培养计划三年行动项目(CIT&TCD201304036
CIT&TCD201404043)
北京市教育委员会科技发展计划项目(KM201410005002)
