电子学报 ›› 2017, Vol. 45 ›› Issue (8): 1911-1918.DOI: 10.3969/j.issn.0372-2112.2017.08.015

• 学术论文 • 上一篇    下一篇

一种基于自适应标记与区域间近邻传播聚类的分水岭图像分割算法

蔡强1,2, 刘亚奇3,4, 曹健1,2, 李海生1,2, 杜军平5   

  1. 1. 北京工商大学计算机与信息工程学院, 北京 100048;
    2. 食品安全大数据技术北京市重点实验室, 北京 100048;
    3. 中国科学院信息工程研究所信息安全国家重点实验室, 北京 100093;
    4. 中国科学院大学网络空间安全学院, 北京 100093;
    5. 北京邮电大学计算机学院, 北京 100876
  • 收稿日期:2015-04-08 修回日期:2016-12-29 出版日期:2017-08-25
    • 通讯作者:
    • 刘亚奇
    • 作者简介:
    • 蔡强,男,1969年出生,重庆市永川市人,博士,教授,主要研究领域为计算机图形学、计算几何、科学可视化、智能信息处理.E-mail:caiq@btbu.edu.cn;曹健,男,1982年出生,山东省临沂市人,博士,副教授,主要研究领域为图像处理、模式识别;李海生,男,1974年出生,山东省德州市人,博士,教授,主要研究领域为计算机图形学、科学可视化、三维模型检索等;杜军平,女,1963年出生,北京市人,博士,教授/博士生导师,北京邮电大学计算机学院,主要研究方向:人工智能、智能信息系统.
    • 基金资助:
    • 国家自然科学基金 (No.61320106006,No.61532006); 北京市自然科学基金 (No.4162019); 北京市科技计划课题 (No.Z161100001616004)

A Watershed Image Segmentation Algorithm Based on Self-adaptive Marking and Interregional Affinity Propagation Clustering

CAI Qiang1,2, LIU Ya-qi3,4, CAO Jian1,2, LI Hai-sheng1,2, DU Jun-ping5   

  1. 1. School of Computer and Information Engineering, Beijing Technology and Business University, Beijing 100048, China;
    2. Beijing Key Laboratory of Big Data Technology for Food Safety, Beijing 100048, China;
    3. State Key Laboratory of Information Security, Institute of Information Engineering, Chinese Academy of Sciences, Beijing 100093, China;
    4. School of Cyber Security, University of Chinese Academy of Sciences, Beijing 100093, China;
    5. School of Computer Science, Beijing University of Posts and Telecommunications, Beijing 100876, China
  • Received:2015-04-08 Revised:2016-12-29 Online:2017-08-25 Published:2017-08-25
    • Supported by:
    • National Natural Science Foundation of China (No.61320106006, No.61532006); National Natural Science Foundation of Beijing Municipality,  China (No.4162019); Beijing Municipal Science and Technology Project (No.Z161100001616004)

摘要: 分水岭算法是一种高效的图像分割算法,能够准确地对图像进行基于区域的分割,但是存在易过分割的问题.为此本文提出一种改进的分水岭算法:首先,对彩色图像进行频谱包络滤波并计算彩色梯度获得梯度图像,再采取一种自适应设定参数的H-minima技术,对梯度图像的极小值区域进行标记;然后,对已标记极小值区域的梯度图像进行分水岭分割;最后,计算分水岭分割所得各区域的颜色矩,作为该区域的颜色特征,并对这些区域进行近邻传播聚类获得分割结果.通过与近年来其它改进的分水岭算法和采用聚类的图像分割算法实验比较,本文所提算法能更加有效地抑制过分割,提高分割准确率,具有良好的自适应性和鲁棒性.

关键词: 分水岭算法, 自适应标记, 近邻传播聚类, 图像分割, 过分割

Abstract: The watershed algorithm can conduct region-based image segmentation effectively and accurately,but it tends to cause over-segmentation.To tackle the above mentioned problem,an improved watershed algorithm is proposed,as follows:first of all,the color gradient is computed using spectrum envelope filtered color image,based on which,regions with minimum gradient are marked using self-adaptive H-minima transformation method.Then,the watershed transform is applied to segment the marked gradient image.Finally,affinity propagation clustering is adopted to merge the regions segmented by the watershed transform,using color moments computed on each local region,to get the final segmentation result.Experiments conducted on public available datasets demonstrate the adaptability and robustness of proposed algorithm,compared with the relative state-of-the-art methods.The proposed method can solve the over-segmentation problem well and get accurate results.

Key words: watershed algorithm, self-adaptive marking, affinity propagation, image segmentation, over-segmentation

中图分类号: