结合闭合解抠图及最小生成树的图论分割算法  

A minimum spanning tree based image segmentation algorithm with closed-form solution

作　　者：王卫星[1,2] 石红玉[1] 

机构地区：[1]福州大学物理与信息工程学院,福州350000 [2]皇家工学院

出　　处：《哈尔滨工业大学学报》2014年第9期123-128,共6页Journal of Harbin Institute of Technology

基　　金：国家自然科学基金资助项目(61170147)

摘　　要：针对图像目标物体与背景边界交错在一起或两者之间边界不明晰以及背景与目标纹理相似的情况,进行图像分割非常困难.为此,提出了一种基于图论（graph theory）及闭合解抠图思想的图像分割算法.首先,利用闭合解抠图算法对图像进行预分割,粗糙地将图像分为前景和背景两部分;其次,提取目标及背景的细节,再分别用改进的图论分割算法细分割目标物体及背景,从而得到最终图像分割结果.实验结果表明,抠图算法避免了前景和背景的混叠,改进的图论算法可有效提高6%~12%的分割精度.与传统的区域合并、通常的图论及阈值算法相比,该算法精度高、效果好,具有显著优越性.For the edges between objects and background in an image are intertwined or their common boundaries are vague as well as the textures of objects and background are similar,a new method based on graph theory and closed-form solution was proposed.First,it uses closed-form solution to initially separate the objects from background roughly,then,to extract the detailed information of inter objects,it applies an improved graph-based algorithm to obtain the final image segmentation results.The test results show that the algorithm of matting avoids aliasing of foreground and background and the improved graph-based algorithm increases segmentation accuracy by 6% ~ 12% effectively.Compared to the traditional algorithms such as region merging,ordinary graph,and thresholding,the new algorithm has the better accuracy and effect,therefore it has the significant superiority.

关 键 词：闭合解抠图 图论 最小生成树 图像分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]