吉林大学学报(理学版) ›› 2018, Vol. 56 ›› Issue (5): 1206-1212.

• 计算机科学 • 上一篇    下一篇

基于Hadoop平台的大数据图像分类机制

张睿萍, 马宗梅   

  1. 中原工学院 计算机系, 郑州 450007
  • 收稿日期:2017-06-14 出版日期:2018-09-26 发布日期:2018-11-22
  • 通讯作者: 张睿萍 E-mail:841482649@qq.com

Large Data Image Classification Mechanism Based on Hadoop Platform#br#

ZHANG Ruiping, MA Zongmei   

  1. Department of Computer Science and Technology, Zhongyuan University of Technology, Zhengzhou 450007, China
  • Received:2017-06-14 Online:2018-09-26 Published:2018-11-22

摘要: 针对大数据图像分类耗时长、 实时性差等问题, 利用云计算技术的优点, 以获得理想的大数据图像分类结果为目标, 提出一种基于Hadoop平台的大数据图像分类机制. 首先收集大量的图像, 构建图像数据库, 并提取图像分类的有效特征; 然后基于Hadoop平台, 采用Map函数对大数据图像分类问题进行细分, 用多节点并行、 分布式地对子问题进行图像分类求解, 得到相应的图像分类结果; 最后利用Reduce函数对子问题的图像分类结果进行组合, 并用VC++6.0编程实现大数据图像分类的仿真模拟测试. 测试结果表明, 该分类机制较好地克服了当前图像分类机制存在的弊端, 大幅度缩短了图像分类的时间, 分类速度可适应大数据图像分类的在线要求, 且图像分类的整体效果明显优于当前其他图像分类机制.

关键词: 图像分类机制, 特征库, 实时性, 分类节点, 特征匹配

Abstract: Aiming at the problem of long timeconsuming and poor realtime of large data images classification, using the advantages of cloud computing technology to obtain the ideal classification results of large data images, we proposed a large data image classification mechanism based on Hadoop platform. Firstly, a large number of images were collected, the image database was constructed, and the effective features of the image classification were extracted. Secondly, based on the Hadoop platform, the Map function was used to subdivide the large data image classification problems, and subproblems were classified and solved by multiple nodes parallel and distributed, and the corresponding image classification results were obtained. Finally, the Reduce function was used to combine image classification results of subproblems, and the simulation test of large data image classification was realized  by using VC++60 programming. Test results show that the proposed classification mechanism can overcome the drawbacks of current image classification mechanism,  greatly shorten  the time of image classification, and the classification speed can adapt to online requirements of large data image classification, and the overall effect of image classification is obviously superior to the other image classification mechanisms.

Key words: image classification mechanism, feature database, realtime, classification node, feature matching

中图分类号: 

  • TP391