【摘要】 Split-Merge算法是图像分割领域的经典算法,思路简洁、运算效率高,但因分裂位置和形状固定,导致出现方块效应和边界定位不准确等问题。现有的各种改进算法不能从根本上解决这些问题,为此本文将基于图的分割算法纳入到Split-Merge框架中来,提出了在Split-Merge框架下进行基于图的分割。新算法把传统的基于图割值的算法作为分裂阶段的处理手段,在合并阶段采用本文提出的一种与访问次序无关的合并算法。新算法既解决了Split-Merge算法中因分裂位置和形状固定而带来的各种问题,也有效地消除了在基于图的分割中常见的"过分割"现象,取得了满意的分割效果。更多还原

【Abstract】 Split-Merge algorithm is a classical algorithm in image segmentation.It is simple and effective.However,because the position and shape of the splitted parts of the image cannot be adaptively adjusted,some problems will arise,such as square effect and inaccuracy of edges. And modifications of the algorithm cannot solve these problems effectively.A graph based image segmentation algorithm in Split-Merge framework is presented.The traditional segmentation based on graph cuts is used to split the image into small regions,and a region merge algorithm is proposed to merge the adjacent regions.The new algorithm effectively solves the problems in Split-Merge algorithm and eliminates over-segmentation in graph based segmentation. Experimental results show that the new method can achieve good results.更多还原