局部颜色模型的交互式Graph-Cut分割算法

Interactive Graph-Cut for image segmentation using local color models

由于目前大多数交互式Graph-Cut分割算法很难达到精确分割且实时交互的效果.对此,提出一种基于局部颜色模型的改进算法.该算法利用Mean-Shift预分割,建立基于局部颜色模型的交互式分割框架,并将像素级的Graph-Cut算法转化为基于区域的算法进行快速求解.预分割之后的区域保持了原有图像的结构,不仅提高了采用局部颜色模型估计分布的准确性,而且基于区域Graph-Cut的算法明显降低了计算的复杂度.实验结果表明,改进后的算法不仅保证了分割的精确性,而且还达到了实时交互.

Most interactive segmentation algorithms based on Graph-Cut do not usually lend themselves to real time applications with accurate segmentation. In this paper, an improved algorithm using local color models was pro- posed to deal with the problem. Using Mean-Shift technology, the proposed algorithm built an interactive image segmentation framework based on local color models. A Graph-Cut algorithm was then applied to the pre-segmented regions instead of image pixels. The pre-segmented regions preserve the image structure, which improves the esti- mation accuracy of distribution based on local color models and dramatically reduces the computational complexity. Experimental results show that the proposed algorithm has good real time interactivity with accurate segmentation.