Image segmentation by directed region subdivision

Abstract

In this paper, an image segmentation method based on directed image region partitioning is proposed. The method consists of two separate stages: a splitting phase followed by a merging phase. The splitting phase starts with an initial coarse triangulation and employs the incremental Delaunay triangulation as a directed image region splitting technique. The triangulation process is accomplished by adding points as vertices one by one into the triangulation. A top-down point selection strategy is proposed for selecting these points in the image domain of grey-value and color images. The merging phase coalesces the oversegmentation, generated by the splitting phase, into homogeneous image regions. Because images might be negatively affected by changes in intensity due to shading or surface orientation change, the authors propose homogeneity criteria which are robust to intensity changes caused by these phenomena for both grey-value and color images. Performance of the image segmentation method has been evaluated by experiments on test images.