Hadoop环境下图像内容检索方法的研究  

Research on the method of retrieval of image content based on Hadoop

作　　者：石祥滨[1,2] 钟刘倍 张德园[1] 

机构地区：[1]沈阳航空航天大学计算机学院,沈阳110136 [2]辽宁大学信息学院,沈阳110036

出　　处：《沈阳航空航天大学学报》2017年第3期63-69,共7页Journal of Shenyang Aerospace University

基　　金：国家自然科学基金(项目编号:61170185;61602320);辽宁省博士启动基金(项目编号:20121034;201601172);辽宁省教育厅科学研究一般项目(项目编号:L2014070;L201607)

摘　　要：近年来,随着互联网技术的发展,图像数据呈现爆炸式增长,图像检索已成为广泛关注的热点问题。提出一种基于Hadoop和Elastic Search的多特征融合图像检索方案。为实现分布检索,该方案给出了视觉词典生成、图片向量化和多维倒排索引的构建方法。为了提高检索精度,设计了Root SIFT、颜色矩、Gabor特征,给出了三种特征融合方法。在Corel-1000标准图像库和ILSVRC2015数据集上的实验结果表明,该方案在分布式平台上具有较快的响应速度以及较高的检索精度。In recent years, with the development of internet technology, the image data shows explosive growth while the image retrieval has become hot in research field. In this paper, an image retrieve scheme based on cloud platform and multi-feature fusion was proposed. In order to achieve distributed parallel re- trieval, the construct methods such as visual dictionary generation, image vectorization and multi-dimensional inverted index were provided by this scheme. To increase the accuracy of retrieval, we applied the RootSIFT feature, color matrix feature and Gabor feature by using a new method to fuse these three features. The ex- perimental results on the Corel-1000 standard image library and the ILSVRC2015 dataset show that the pro- posed scheme can provide fast response speed and high retrieval precision on the distributed platform.

关 键 词：云平台 特征融合 CBIR 分布式计算 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]