王立 陈军峰



摘 要: 传统基于内容的图像检索方法通过相似度测量算法获取检索结果,对海量图像存在检索效率低和精度差的弊端,因此设计基于Hadoop分布式的海量图像检索方法,其基于Hadoop云平台对海量数码图像实施分布式运算,采集图像SURF特征,采用K?Means聚类方法将相似图像SURF特征聚集起来,通过TF?IDF数据挖掘技术对图像特征实施量化,进而基于Hadoop平台中的Lucene框架塑造海量图像数据的索引模块和搜索模块,依据用户输入的图像SURF特征塑造海量图像数据索引,完成相似图像的准确检索。实验结果说明,所提图像检索方法检索出的图像质量佳,对海量图像进行检索的效率和精度高。
关键词: Hadoop分布式; 海量图像; SURF特征; K?Means聚类; 检索; 数据挖掘
中图分类号: TN911.73?34; TP391 文献标识码: A 文章编号: 1004?373X(2018)09?0062?06
Abstract: The traditional content based image retrieval method obtains the retrieval results by means of similarity measurement algorithm, which has the disadvantages of poor retrieval accuracy and low retrieval efficiency for massive image. Therefore, a massive image retrieval method based on Hadoop distribution was designed to implement the distributed computing for massive digital image on the basis of its Hadoop cloud platform. The image SURF feature is acquired. And then the K?Means clustering method is used to assemble the SURF feature of similar images together. The TF?IDF data mining technology is used to quantify the image features, and then the index module and search module of massive image data are constructed on the basis of Lucene framework in Hadoop platform. According to the image SURF feature of user input, the data index of massive image was constructed to retrieve the similar images accurately. The experimental results show that the image retrieval method has high retrieval image quality, and high retrieval efficiency and accuracy of massive image.
Keywords: Hadoop distribution; massive image; SURF feature; K?Means clustering; retrieval; data mining
0 引 言
由于当前移动互联网的不断发展和网络服务应用领域的逐渐扩展,产生的用户数据量也不断增长。图像拥有详细、直观的属性,是重要的多媒体数据载体。图像检索方法能够从海量图像数据内采集同目标图像相似度最高的特征以及内容的图像,为用户检索到满足自身兴趣度的图像。随着图像数据量的提升,海量数据的图像检索问题成为相关学者研究的热点。传统基于内容的图像检索方法,通过相似度测量算法获取检索结果,对海量图像数据处理能力差,存在图像检索效率低和精度差的弊端。
面对该问题,本文提出基于Hadoop分布式的海量图像检索方法,提高海量图像检索的效率和精度,增强图像检索质量。……