• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2013, Vol. 35 ›› Issue (5): 118-123.

• 论文 • 上一篇    下一篇

基于K-均值聚类及二叉树决策的图像去噪

刘永霞1,张朝晖1,2,张艳敏1   

  1. (1.河北师范大学数学与信息科学学院,河北 石家庄 050024;2.河北省计算数学与应用重点实验室,河北 石家庄 050024)
  • 收稿日期:2011-12-23 修回日期:2012-05-21 出版日期:2013-05-25 发布日期:2013-05-25
  • 基金资助:

    国家自然科学基金天元数学基金资助项目(10926179);河北省科学技术重大支撑计划资助项目(10243554D);河北省科学技术研究与发展计划资助项目(072435158D;09213515D;09213575D)

Image denoising based on K-Means
clustering and binary tree decision        

LIU Yongxia1,ZHANG Zhaohui1,2,ZHANG Yanmin1   

  1. (1.College of Mathematics and Information Science,Hebei Normal University,Shijiazhuang 050024;
    2.Key Laboratory of Computational Mathematics and Applications of Hebei Province,Shijiazhuang 050024,China)
  • Received:2011-12-23 Revised:2012-05-21 Online:2013-05-25 Published:2013-05-25

摘要:

针对图像中椒盐噪声的抑制,提出了一种新的滤波算法。算法首先借助K均值聚类将当前像素所在邻域的灰度分布进行有效划分;然后,构建噪声污染像素识别规则,借助多层二叉树决策实现不同类型噪声污染像素的检测。算法只针对噪声污染像素进行自适应滤波,而不改变非污染像素的取值。实验表明,本文算法在有效抑制噪声的同时可较好保留图像的细节等有用信息;对于噪声污染严重的图像,本算法明显优于传统中值滤波及文献[7]的算法。

关键词: 椒盐噪声, K均值聚类, 二叉树, 图像滤波, 噪声检测

Abstract:

In this paper, a new filter algorithm was proposed for pepperandsalt noise suppression. Firstly, the neighborhood of each given pixel is partitioned by Kmeans clustering according to the local grey level distribution. Secondly, the recognition rules for noisepolluted pixel detection are constructed, and the noise pixel can be detected based on multilayer binary tree decision. The proposed algorithm only filters the recognized noise pixels without changing those nonpolluted pixel values. Experimental results show that the proposed algorithm can efficiently preserve informative details when filtering image noise. For those images with strong noise pollution, the proposed algorithm outperforms both median filter and the algorithm proposed in[7].

Key words: pepper-and-salt noise;K-means clustering;binary tree;image filtering;noise detection