Abstract: The existing interactive image segmentation algorithms are not efficient enough to process high resolution images. In order to solve this problem, this paper proposes an efficient algorithm for interactive image segmentation based on simple linear iterative clustering (SLIC) and Delaunay graph cut. Firstly, a kind of simplified but efficient SLIC superpixel method is used to group pixels into perceptually meaningful atomic regions, so the representative pixels can be extracted from these regions. Secondly, Delaunay triangulation is adopted for graph construction on the representative pixels in the bounding box. Finally, min-cut/max-flow algorithm is applied to divide the nodes in the graph into two parts, which represent the foreground and background in the image. The experimental results and comparisons with other interactive image segmentation algorithms demonstrate the high efficiency and accuracy of the proposed algorithm.

Key words: image segmentation, simple linear iterative clustering (SLIC), Delaunay triangulation, min-cut/max-flow

摘要： 针对现有的交互式图像分割算法在处理高分辨率图像时仍不够高效的问题，提出了一种基于简单线性迭代聚类（simple linear iterative clustering，SLIC）与Delaunay图割的交互式图像分割算法。使用一种简化但是高效的SLIC算法将图像分割为多个在感知上有意义的原子区域，并提取这些区域的代表像素点；对处在背景矩形框内的代表像素点进行Delaunay三角剖分，构建图结构；最后利用最小割最大流算法将图中的节点分为两部分，并将这些节点对应为相应的原子区域，达到将图像分割为前景和背景的目的。与其他交互式图像分割算法进行实验对比，结果表明所提算法在计算效率上有较大提升，并更为准确。

关键词: 图像分割, 简单线性迭代聚类（SLIC）, Delaunay三角剖分, 最小割最大流