Image segmentation by connectivity preserving relinking in hierarchical graph structures

摘要

The method of image segmentation by pyramid relinking is extended to the formalism of hierarchies of region adjacency graphs. This approach has a number of advantages: (1) resulting regions are connected; (2) the method is adaptive, and therefore artifacts caused by a regular grid are avoided; and (3) information on regions and boundaries between regions can be combined to guide the segmentation procedure. The method is evaluated by the segmentation of a number of synthetic and natural images. Note: This research was supported by the Foundation for Computer Science in the Netherlands (SION) with financial support from the Netherlands Organization for Scientific Research (NWO). This research was part of a project in which the TNO Institute for Human Factors, CWI and the University of Amsterdam cooperate. Most of this work was performed while the author was a guest at the Technical University of Vienna, Austria, in the framework of the Erasmus exchange program of the European Community.