基于骨架层次分解的目标的图表示

基于骨架的目标表示技术是模式识别和计算机视觉的重要研究内容，近年来人们提出了许多骨架化算法，但是有关利用骨架信息表示并识别目标的研究还非常有限。Ablameyko等1996年提出了通过分解由距离标号的骨架为有意义的结构基元从而获得目标的层次结构图的方法。该图可以准确地刻画基元之间的拓扑关系，但是它对于骨架中的噪声比较敏感。主要表现为噪声基元破坏其它基元的完整性和图的稳定性。该文采用将分支编组为分支链以及构造多尺度结构图的改进策略来克服这些缺点，最终获得了目标的节点数更小、节点显著度更高、节点间连接关系更稳定的多尺度图，从而显著地提高后续利用不精确图匹配技术进行目标识别的效率。这项技术已经被应用于一个基于形状特征的图像数据库检索系统中。

Graph Representation for Objects Based on Hierarchical Decomposition of Skeletons

Skeleton based object representation techniques are of importance in pattern recognition and computer vision. Many skeletonization algorithms have been proposed in recent years. However the studies on how to use skeleton information to represent and recognize objects are very limited. Ablameyko et al. [1] presented an algorithm to construct the hierarchical structure graph of the object by decomposition of the distance labeled skeleton into its meaningful structure elements. This graph can exactly describe the topological relationship of its structure elements. However it is sensitive to the noise in the skeleton, for example, the noise elements can destroy the integrality of the other elements and the stability of the graph. In order to overcome these drawbacks, an improvement strategy is presented in this paper which employs a skeleton branches grouping procedure and a multi scale hierarchical structure graph constructing procedure. It has been shown that the resultant graph has fewer and more notable nodes and its structure is more stable than before. This graph can greatly improve the efficiency of the following inexact graph matching procedure. This technique has been applied in a shape feature based image database retrieval system.