Abstract: The shortages of conventional ultrametric contour map (UCM) algorithm are weak-adapted for contours, weak-adapted for hierarchies matching and lots of fragments for segmentation. To solve the shortages, this paper proposes a novel improved hierarchical image segmentation algorithm based on self-adapted objects and context matching strategy. First, the algorithm uses “edge boxes” to extract the key contours, and then uses the weighted watershed algorithm to combine contour regions, and the contours are transferred for UCM hierarchical trees. Then, a dynamic programming method is used for self-adapted objects and context matching. At last, the re-scaling UCM hierarchical trees are used for image segmentation. Experimental results on BSDS500 datasets show that the proposed algorithm has a significant improvement on image segmentation. Three measure indexes are segment cover (SC), probabilistic region index (PRI) and information variation (IV) which have achieved the best results on optimal dataset scale (ODS) and optimal image scale (OIS). The re-scaling UCM hierarchical trees satisfy the same hierarchies are segmented for the same objects, and reduce the fragments of image segmentation. The proposed algorithm is outperformed than the state-of-the-art algorithms for image segmentation, and satisfies the time comple-xity in a certain stage.

Key words: edge boxes, weighted watershed, ultrametric contour map, dynamic programming, self-adapted matching

摘要： 针对超度量轮廓图（ultrametric contour map，UCM）层级图像分割算法对轮廓适应性弱、层级匹配能力较弱且分割碎片较多等问题，提出了一种自适应目标与内容匹配的改进UCM层级图像分割算法。该算法首先使用“轮廓盒子”提取图像关键轮廓，然后使用加权分水岭算法合并区域，提升轮廓适应性，并产生UCM层级树；随后，采用动态规划的方式自适应完成目标与内容匹配，最后使用调整尺度后的UCM层级树完成图像分割。在BSDS500数据集上进行了分割实验，实验结果表明提出的算法在各项分割指标上获得了显著的提升。分割掩盖率（segment cover，SC）、概率边缘指标（probabilistic region index，PRI）和信息变化率（infor-mation variation，IV）三个衡量指标分别在最优数据集尺度（optimal dataset scale，ODS）和最优图像尺度（optimal image scale，OIS）上获得了最佳的效果。UCM层级树通过尺度的调整，能够保证相同尺度的层级分割为同一层，减少了分割碎片，保证了层级匹配。该算法在分割精度上超越了当前大多数主流图像分割算法，同时保证时间复杂度在同一个级别。

关键词: 轮廓盒子, 加权分水岭算法, 超度量轮廓图算法, 动态规划, 自适应匹配