一种改进的Graph Cuts交互图像分割方法  

An Improved Interactive Graph Cuts for Image Segmentation

作　　者：章卫祥[1] 周秉锋[1] 

机构地区：[1]北京大学计算机科学技术研究所,北京100871

出　　处：《影像技术》2007年第4期22-24,共3页Image Technology

摘　　要：本文提出了一种改进的Graph Cuts交互图像分割方法。Graph Cuts交互图像分割方法〔1〕首先由用户选定部分像素作为对象和背景,其余像素为未知区域;然后根据以像素为顶点,以像素相邻关系为边,构造一个图;最后通过图的最小分割方法将图像分为对象和背景两部分。此方法分割图像的结果直接受到用户选定对象和背景像素操作的影响,对象和背景边界的像素容易被分割错误。我们分别对对象区域和背景区域进行腐蚀操作,使分割错误的像素重新变为未知区域(对于在对象或背景内部被错误划分的像素,可以利用类似画笔的工具,直接将其标为对像或背景),然后重新进行一次Graph Cuts分割。由于这次选定了大部分的对象和背景区域,实验结果表明,最后分割结果正确率明显提高了。In this paper, we have improved the interactive graph cuts method for image segmentation. There are three steps in the interactive graph cuts method. First, some pixels are selected as the object and background seeds, and the left is the unknown region. Secondly, a graph is constructed, in which the vertices are all the pixels and the edges are the neighborhood relationship between the pixels. Last, the image is cut into two parts, the object and background, using the min-cut algorithm. The result is directly affected by the seeds which are selected by the user, and the pixels on the border of the object may be partitioned wrongly. We erode the object and background region to exclude the wrong pixels, then we use those as object and background seeds and do the graph cuts again. The experiments show that the result is more accurate because most pixels of the object and background are selected as seeds.

关 键 词：计算机图像处理 图像分割 GRAPH CUTS 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]