电子学报 ›› 2018, Vol. 46 ›› Issue (10): 2519-2526.DOI: 10.3969/j.issn.0372-2112.2018.10.028

• 科研通信 • 上一篇    下一篇

一种基于二维局部二值模式的纹理图像分类方法

王凯丽, 张艳红, 肖斌, 李伟生   

  1. 重庆邮电大学计算智能重点实验室, 重庆 400065
  • 收稿日期:2017-04-20 修回日期:2017-11-23 出版日期:2018-10-25 发布日期:2018-10-25
  • 通讯作者: 肖斌
  • 作者简介:王凯丽,女,1992年生于河南商丘.硕士研究生在读,研究方向为基于局部二值模式的纹理识别;张艳红,女,1991年生于河南焦作.硕士,研究方向为数字图像分析和模式识别;李伟生,男,1975年生.教授,研究方向为智能信息处理与模式识别.
  • 基金资助:
    国家自然科学基金(No.61572092);国家自然科学基金—广东联合基金(No.U1401252);国家重点研发计划(No.2016YFC1000307-3)

Texture Images Classification Based on Two Dimensional Local Binary Patterns

WANG Kai-li, ZHANG Yan-hong, XIAO Bin, LI Wei-sheng   

  1. Chongqing Key Laboratory of Computation Intelligence, Chongqing University of Posts and Telecommunication, Chongqing 400065, China
  • Received:2017-04-20 Revised:2017-11-23 Online:2018-10-25 Published:2018-10-25

摘要: 局部二值模式(Local Binary Pattern,LBP)在纹理分类中受到越来越多的关注,传统的基于局部二值模式的图像识别方法在LBP直方图统计时仅仅考虑到LBP模式值本身的数量统计,却忽略了模式值之间的相关性.针对这一问题,本文提出一种二维局部二值模式(Two Dimensional Local Binary Pattern,2DLBP)方法,并用于纹理图像识别.首先以旋转不变均匀LBP特征图为基础,引入滑动窗口和LBP模式对的概念,统计LBP模式图的上下文信息,构造出2DLBP特征;然后改变LBP中的半径参数,构造图像的多分辨率2DLBP特征,并利用支持向量机(SVM)的分类方法进行纹理分类;最后选取Brodatz、CUReT、UIUC、FMD四个公开纹理库分别进行纹理分类测试.理论验证表明该方法具有良好的通用性,可以与LBP的其他变型结合成为新的图像特征构造方法.同时,实验结果表明,本文提出方法具有较好的纹理图像分类能力.

关键词: 局部二值模式(LBP), 纹理图像, 上下文信息, 纹理图像分类

Abstract: Local binary patterns have been widely used in texture images classification.However,conventional LBP methods focus on the distribution of LBP values and ignore the spatial contextual information between LBP patterns.In this paper,a texture images classification method based on two-dimensional Local Binary Pattern (2DLBP) is proposed.The proposed method introduces a sliding window to count the weighted occurrence number of LBP pairs on the feature map of rotation invariant uniform LBP.The radius of LBP is also changed to obtain the multi-resolution 2DLBP features.At last,texture images are classified using the methods of the support vector machine (SVM).Theoretical validation shows that the proposed method is a generalized framework,and can be integrated with other LBP variants to derive a new feature extraction method.Experimental results show that,compared with the conventional LBP,the variants of LBP,and some state of the art texture classification methods,the proposed method achieves acceptable performance in texture images classification.

Key words: local binary pattern, texture images, contextual information, texture image classification

中图分类号: