基于图割的交互式彩色图像目标提取方法

Interactive color image object extraction method based on graph cuts

摘要:

根据图像边缘基本性质,运用图割理论,把目标 提取问题转化为网络最大流-最小割问题,提出一种人机交互的彩色图像目标提取方法。根 据目标形状及大小,调节触笔笔尖粗 细,在目标区域内较为随意地划一道或若干道连通的粗线,以该粗线边界作为初始活动轮廓 线,在张力作用下向外侧膨胀生成环状域,对其构造s-t网络, 进行最小代价切割获得新的 活动轮廓线。由此经过若干次变形迭代,活动轮廓线最终收敛于目标边界。当因目标边界模 糊不清等原因导致提取结果局部出现错误时,算法提供方便快捷、安全导向、手自结合的纠 错方法。实验表明:当目标边界呈现较为清晰时,本文算法以人能够接受的毫秒级时间实时 快速作出响应,目标提取结果正确率在95%左右。当提取结果出现局 部错误时,算法通过人机结合与交互,对错误进行有效纠正。

Abstract:

According to the basic character of image edge and the theory of graph cuts,the problem of object extraction is transformed into the problem of maximu m flow-minimum cut in s-t network.And a method of obje ct extraction in color im age based on human-computer interaction is proposed.According to the shape and size of the object,the thickness of the stylus tip is adjusted,and one or seve ral connected lines are delineated randomly in the target area.The rough line b oundary is used as the initial active contour,which expands to the outside unde r the action of tension to form an annular region.The s-t network is construct e d and the new active contour is obtained by cutting it at the minimum cost.Afte r several deformation iterations,the active contour finally converges to the ta rget boundary.When local errors occur in the extraction results due to ambiguou s boundary of the object,the algorithm provides a convenient,fast and safe-or i ented error correction method in automatic or fully manual way.Experiments show that when the boundary of the target is certain clear,the algorithm can respon d quickly in real time within a second,which is acceptable to human.The accura cy of the object extraction results is about 95%.When local errors occur in the extraction results,the algorithm corrects the errors effectively through the i nteraction and combination of human and computer.