Depth-First-Search Based Region Merging for the Waterfall

摘要

Watershed transformation is a popular method to segment regions in an image. After being proposed in 70's, it has been evolved in various ways. However, the problem of over-segmentation still remains because the existing approaches are computationally expensive, they apply a waterfall technique to the result of the watershed transformation, merging similar regions. The merging requires navigating the graph representing regions and their neighboring relationship. In this paper, we propose to improve navigation and merging process by transforming the graph into a tree and using the depth-first-search. The proposed method was able to simplify the region merging process while producing similar results.