一种基于Multiway cut的多对象图像分割  

Multi-object segmentation based on Multiway cut

作　　者：孙统风[1,2] 丁世飞[2] 任子晖[1] 

机构地区：[1]中国矿业大学信息与电气工程学院,江苏徐州221116 [2]中国矿业大学计算机科学与技术学院,江苏徐州221116

出　　处：《计算机应用研究》2010年第8期3138-3140,3149,共4页Application Research of Computers

基　　金：国家"863"计划资助项目(2006AA01Z128);江苏省自然科学基金资助项目(BK2009093);中国矿业大学科技基金资助项目(2007A047)

摘　　要：多对象分割是图像处理中的一个难题,基于Multiway cut的图像分割是一种人工交互式多对象分割方法,能够实现图像的粗分割和精确分割。使用分水岭分割图像,把图像分割为属性相似的小区域;根据交互建立节点层次图,构建带权无向网络;不同层次的节点参与不同的运算,采用Multiway cut迭代分割;交互和分割可以多次执行,直至满足用户的要求。实验结果表明,该方法人工参与方便,准确度得到提高,速度满足现场操作的要求。Multi-object segmentation is a difficult problem in image processing. Image segmentation,based on Multiway cut, is an interactive multi-object image segmentation,which can implement segmentation from coarseness to accurateness. Firstly, used watershed to segment a image into small regions with the same attributes. Then built a node hierarchy and a weighted undirected network based on interactions. Used the nodes of different hierarchy play different roles and the algorithm of Multiway cut to segment image iteratively. Interaction and segmentation could be executed many times until user’s requirements were met. Experimental results show that the method can reduce manual participation,effectively improve accuracy and its speed can meet the requirements of field operations.

关 键 词：图像分割 分水岭 图割 Multiway CUT 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]