基于改进分水岭和区域合并的彩色图像分割  

Color image segmentation based on improved marked watershed transform and region merging

作　　者：李占波[1] 杨二伟 李进文[1] 

机构地区：[1]郑州大学信息工程学院,河南郑州450001

出　　处：《计算机工程与设计》2014年第7期2458-2461,2503,共5页Computer Engineering and Design

摘　　要：提出一种基于改进分水岭变换和区域合并的彩色图像分割算法。采用相位一致性模型进行边缘提取,将提取结果进行标记分水岭变换;应用提出的基于扩展邻域和无量纲阈值的区域合并算法进行同质区域合并,克服了传统区域合并算法中奇异区域影响同质区域联通的局限性。实验结果表明,该算法适用于多数彩色图像,能得到与人眼视觉判断基本一致的分割效果。A hybrid color image segmentation algorithm combining marked watershed transform and improved region merging was proposed.Phase consistency model was used for edge extraction whose result was used as input of marked watershed algo-rithm.Finally,homogeneous regions was merged using proposed region merge algorithm which was based on extended neighbor-hoods and dimensionless threshold.The proposed algorithm overcame the limitations of traditional region merging algorithm that singular regions might destroy connectivity of homogeneous regions.Experimental results showed that the new algorithm could be applied to most color images and the results were consistent with human visual judgments.

关 键 词：相位一致性 标记分水岭 扩展邻域 区域合并 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]