基于多阈值融合的图像分割

Multi-Threshold Fusion Based Image Segmentation

提出了一种基于知识的多阈值融合图像分割新方法 .首先利用一组多阈值分割结果建立连通域生长树 .然后判断树叉对应的连通域合并是否合理 ,为此提出了连通体元、体元生命期、体元体积等概念 ,结合灰度均匀性定义出通用合并准则 .最后将图像各位置的最佳连通域组合为最终图像分割结果 .该算法充分利用了目标的灰度和空间属性 ,对灰度平稳和渐近变化的多目标图像分割非常有效 .此外 ,该算法可以有效融合具体应用的先验知识 ,具有很高的智能性 .

A knowledge-based image segmentation with multi-threshold fusion is presented. First, the connected-region-growing tree is established from a set of segmented images with different thresholds. Then the merges of connected regions corresponding to the tree crotches are checked for rationality, for which we introduce the concept of primitive connection volume, whose volume and lifetime are also defined, which are used to define the merging rules between connected regions, together with the traditional gray uniformity criteria. The optimal connected regions for each position are grouped to form the segmentation result. This algorithm combines the gray-level and spatial properties of objects in images, and is very effective for multi-object image segmentation, which obeys the assumption of gray uniformity and smoothness. Besides, it is ready for combining a prior knowledge of specific applications, which makes itself intelligent.