北京航空航天大学学报 ›› 2015, Vol. 41 ›› Issue (2): 302-310.doi: 10.13700/j.bh.1001-5965.2014.0471

• 论文 • 上一篇    下一篇

基于朴素贝叶斯K近邻的快速图像分类算法

张旭1,2, 蒋建国1, 洪日昌1, 杜跃2   

  1. 1. 合肥工业大学 计算机与信息学院, 合肥 230009;
    2. 陆军军官学院 计算机教研室, 合肥 230031
  • 收稿日期:2014-04-28 出版日期:2015-02-20 发布日期:2015-03-12
  • 通讯作者: 洪日昌(1981—), 男, 安徽黄山人, 教授, hongrc@hfut.edu.cn, 主要研究方向为多媒体内容分析、信息检索、多媒体问答. E-mail:hongrc@hfut.edu.cn
  • 作者简介:张旭(1981—), 男, 安徽亳州人, 博士生, zhangxu21cn@163.com
  • 基金资助:

    国家自然科学基金资助项目(61172164)

Accelerated image classification algorithm based on naive Bayes K-nearest neighbor

ZHANG Xu1,2, JIANG Jianguo1, HONG Richang1, DU Yue2   

  1. 1. School of Computer and Information, Hefei University of Technology, Hefei 230009, China;
    2. Department of Computer, Army Officer Academy of PLA, Hefei 230031, China
  • Received:2014-04-28 Online:2015-02-20 Published:2015-03-12

摘要:

朴素贝叶斯最近邻(NBNN)分类算法具有非特征量化和图像-类别度量方式的优点,但算法运行速度较慢,分类正确率较低.针对此问题,提出一种朴素贝叶斯K近邻分类算法,基于快速近似最近邻(FLANN)搜索特征的K近邻用于分类决策并去除背景信息对分类性能的影响;为了进一步提高算法的运行速度及减少算法的内存开销,采用特征选择的方式分别减少测试图像和训练图像集的特征数目,并尝试同时减少测试图像和训练图像集中的特征数目平衡分类正确率与分类时间之间的矛盾.该算法保留了原始NBNN算法的优点,无需参数学习的过程,实验结果验证了算法的正确性和有效性.

关键词: 图像分类, 最近邻, K近邻, 图像-类别距离, 特征选择

Abstract:

Naive Bayes nearest neighbor (NBNN) classification algorithm possesses merits of avoiding feature quantization and image-to-class distance measurement, but it faces limitation of slow speed and low classification accuracy. To address the problem, a naive Bayes K-nearest neighbor classification algorithm was presented, where K-nearest neighbor searched by fast library for approximate nearest neighbors(FLANN) was employed and the influence of background information was removed. In order to improve the running speed and reduce memory cost, feature selection was included for reducing feature number of test and training images. And an attempt was tried to balance the contradictory between classification accuracy and classification time by reducing feature number of test image and training images simultaneously. The algorithm retains merits of original NBNN algorithm and requires no parameter learning process. Experimental results verify the correctness and effectiveness of the algorithm.

Key words: image classification, nearest neighbor, K nearest neighbor, image-to-class distance, feature selection

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发