计算机工程与应用 ›› 2009, Vol. 45 ›› Issue (4): 202-204.DOI: 10.3778/j.issn.1002-8331.2009.04.058

• 图形、图像、模式识别 • 上一篇    下一篇

应用谱直方图和EM的纹理图像分割算法

那 婕   

  1. 辽宁师范大学 计算机与信息技术学院,辽宁 大连 116029
  • 收稿日期:2008-05-06 修回日期:2008-08-12 出版日期:2009-02-01 发布日期:2009-02-01
  • 通讯作者: 那 婕

Spectral histograms and EM-based texture image segmentation algorithm

NA Jie   

  1. School of Computer and Information Technology,Liaoning Normal University,Dalian,Liaoning 116029,China
  • Received:2008-05-06 Revised:2008-08-12 Online:2009-02-01 Published:2009-02-01
  • Contact: NA Jie

摘要: 提出了一种基于谱直方图和EM的纹理图像分割算法。为了获得图像的纹理特征,首先对原始图像进行滤波,然后利用谱直方图的思想和方法,把每个图像子块用独立的谱直方图来进行表示,该直方图在图像的表示上具有很好的本质扩展性。其次采用?字2作为距离度量函数对谱直方图进行计算得到特征值。为了得到初始分割结果,通过EM(期望最大化)对特征值矩阵进行分类,得到有效的初始分割结果。最后使用形态学方法对边界进行定位,从而实现图像的由粗到细的分割。实验结果证明:该算法用于纹理图像分割能获得很好的效果。

关键词: 谱直方图, 纹理分割, 距离函数, 期望最大化算法, 形态学

Abstract: A texture image segmentation algorithm based on spectral histograms and Expectation Maximization(EM) is recommended.First of all,to acquire the texture features,using the algorithm of filter selection,then using spectral histogram’s method and idea,each subimage is presented on the spectral histograms representation.Spectral histograms have intrinsic generalization of image representations.For another,using ?字2 as the distance measure function to calculate the feature statistic of spectral histogram.To derive the initial segmentation result,feature statistic matrix is used as the feature vector to apply expectation maximization image segmentation to achieve effective segmentation.Finally,using the method of morphology to localization of region boundaries.Experimental results demonstrate that high segmentation accuracy can be achieved with the proposed methods.

Key words: spectral histograms, texture segmentation, distance function, Expectation Maximization(EM) algorithm, morphology