东北大学学报:自然科学版 ›› 2018, Vol. 39 ›› Issue (4): 483-487.DOI: 10.12068/j.issn.1005-3026.2018.04.006

• 信息与控制 • 上一篇    下一篇

改进的局部扩展拟合图像分割方法

宫照煊1,2, 覃文军1, 郭薇2, 赵大哲1,3   

  1. (1. 东北大学 计算机科学与工程学院, 辽宁 沈阳110169; 2. 沈阳航空航天大学 计算机学院, 辽宁 沈阳110036; 3. 东软集团股份有限公司, 辽宁 沈阳110169)
  • 收稿日期:2016-11-02 修回日期:2016-11-02 出版日期:2018-04-15 发布日期:2018-04-10
  • 通讯作者: 宫照煊
  • 作者简介:宫照煊(1985-),男,辽宁沈阳人,东北大学博士研究生; 赵大哲(1960-),女,辽宁沈阳人,东北大学教授,博士生导师.
  • 基金资助:
    国家自然科学基金资助项目(51171041).国家高技术研究发展计划项目(2015AA020106); 中央高校基本科研业务费专项资金资助项目(N150408001, N140407001); 国家自然科学基金青年基金资助项目(61402298); 国家自然科学基金资助项目(61373088).

Improved Region-Scalable Fitting Image Segmentation Method

GONG Zhao-xuan1,2, TAN Wen-jun1, GUO Wei2, ZHAO Da-zhe1,3   

  1. 1. School of Computer Science & Engineering, Northeastern University, Shenyang 110169, China; 2.School of Computing, Shenyang Aerospace University, Shenyang 110036, China; 3. Neusoft Group Co., Ltd., Shenyang 110169, China.
  • Received:2016-11-02 Revised:2016-11-02 Online:2018-04-15 Published:2018-04-10
  • Contact: ZHAO Da-zhe
  • About author:-
  • Supported by:
    -

摘要: 局部扩展拟合(RSF)模型可以有效解决灰度不均匀的图像分割问题,但传统的局部拓展拟合模型只考虑了图像上一点在其局部拓展区域内与零水平集相交的情况,易导致过分割现象.为此, 提出了一种改进的局部拓展拟合方法,即在RSF模型的基础上考虑图像上每一点在其扩展邻域内与零水平集不相交的情况,并重新定义了此情况下灰度能量的取值,使该点的内部灰度能量与外部灰度能量严格相等.实验结果表明所提方法有效解决了由初始化问题导致的过分割现象.

关键词: 局部扩展拟合, 灰度不均匀, 分割, 初始化, 零水平集

Abstract: The region-scalable fitting(RSF) model can effectively solve the problem of intensity inhomogeneities, but the traditional RSF model only considers the case where the point on the image intersects with the zero level set in its local extension area, which easily leads to over-segmentation. To solve the problem, an improved region-scalable fitting method was proposed, i.e., the case where every point in the image does not intersect with the zero level set in its extended neighborhood was considered based on the RSF model, and the value of intensity energy in this case was redefined to make the inner intensity energy and the outer intensity energy strictly equal. Experiment results showed the proposed method effectively solves the over-segmentation caused by the initialization problem.

Key words: region-scalable fitting, intensity inhomogeneities, segmentation, initialization, zero level set

中图分类号: