作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2018, Vol. 44 ›› Issue (7): 230-236,243. doi: 10.19678/j.issn.1000-3428.0047158

• 图形图像处理 • 上一篇    下一篇

基于显著性的Grabcut图像分割方法

王万耀,段先华,徐丹,於跃成,黄炜亮   

  1. 江苏科技大学 计算机科学与工程学院,江苏 镇江 212000
  • 收稿日期:2017-05-11 出版日期:2018-07-15 发布日期:2018-07-15
  • 作者简介:王万耀(1992—),男,硕士研究生,主研方向为图像处理;段先华,教授、博士;徐丹,讲师、博士;於跃成,副教授、博士;黄炜亮,硕士研究生。
  • 基金资助:

    国家自然科学基金(61305058);江苏省高校自然科学研究面上项目(16KJB52009,15KJB520008)。

Grabcut Image Segmentation Method Based on Saliency

WANG Wanyao,DUAN Xianhua,XU Dan,YU Yuecheng,HUANG Weiliang   

  1. School of Computer Science and Engineering,Jiangsu University of Science and Technology,Zhenjiang,Jiangsu 212000,China
  • Received:2017-05-11 Online:2018-07-15 Published:2018-07-15

摘要:

传统Grabcut算法需要人工交互,在图像前景和背景颜色相似时容易产生错误分割的问题。为此,结合显著性测量和Grabcut算法,提出一种新的图像分割方法。改良显著性的度量方法以获得较好的显著图,进而将显著图转换成约束项加入到Grabcut的区域项中,从而提高区域 项的可靠性,利用显著图代替人工交互初始化Grabcut,实现Grabcut自动分割,并减少其在图像前景和背景颜色相似时的错误分割。实验结果表明,与传统的Grabcut算法和显著性检测方法相比,该方法可以获得较好的图像分割结果。

关键词: 显著性, Grabcut算法, 图像分割, 视觉注意机制, 高斯混合模型

Abstract:

Traditional Grabcut algorithm needs human interaction,which produces segmentation errors when the colors of the foreground and background are similar.To overcome this disadvantage,by combining Grabcut and saliency measurement,a novel image segmentation algorithm is proposed.An improved saliency measure is developed to obtain better saliency maps.To improve the reliability,these saliency maps are transformed into constraints that are added to the Grabcut regions.The saliency maps,instead of the artificial frames,are used to initialize the Grabcut,with the aim of realizing the automatic segmentation and reducing the segmentation error when the foreground and background colors are similar.Experimental results show that compared with the traditional Grabcut algorithm and the saliency detection algorithm,the proposed algorithm can obtain better image segmentation results.

Key words: saliency, Grabcut algorithm, image segmentation, visual attention mechanism, Gaussian mixture model

中图分类号: