Computer Engineering and Applications ›› 2009, Vol. 45 ›› Issue (22): 156-158.DOI: 10.3778/j.issn.1002-8331.2009.22.051

• 图形、图像、模式识别 • Previous Articles     Next Articles

Color image segmentation method combining mean shift and region merging

PAN Hong-yan   

  1. School of Computer Science and Engineering,Wenzhou University,Wenzhou,Zhejiang 325000,China
  • Received:2008-04-29 Revised:2008-07-28 Online:2009-08-01 Published:2009-08-01
  • Contact: PAN Hong-yan



  1. 温州大学 计算机科学与工程学院,浙江 温州 325000
  • 通讯作者: 潘红艳

Abstract: Color image segmentation is a critical step from color image processing to analysis.In HSV feature space,a color image segmentation algorithm using fused information of color and space features is proposed by combining mean-shift and region merging.First,this paper uses an ameliorated mean shift method to gain local minimums,namely the cluster centers,with the bandwidths and weights acquired on the basis of a better self-adaptive principle according to global information,thereby making the algorithm more applicable;then,the over-segmentation of texture and uneven illumination caused is settled by region merging based on fisher distance.Region merging is ended according to divergence principle.The results demonstrate that the method is effective and optimistic.

Key words: color image segmentation, self-adaptive mean shift, fisher distance, region merging

摘要: 彩色图像分割是从图像处理到分析的关键步骤之一。结合了均值漂移和区域合并算法,在HSV空间,提出一种融合颜色和空间信息的彩色图像分割方法。该算法先由改进的均值漂移(Mean shift)算法求取各局部极值(聚类中心),并利用全局信息,改进了现有的带宽求取和权重设置自适应法则;针对均值漂移带来的纹理和光影的过分割,使用改进的fisher距离进行区域合并,取散度作为停止度量。实验证明,此算法是有效的。

关键词: 彩色图像分割, 自适应均值漂移, fisher距离, 区域合并