Local Watershed Operators for Image Segmentation

In this paper, we propose local watershed operators for the segmentation of medical structures. Watershed transform is a powerful technique to partition an image into many regions while retaining edge information very well. Most watershed algorithms have been designed to operate on the whole or cropped image, making them very slow for large data sets. In this paper, we propose a computationally efficient local implementation of watersheds. In addition, we show that this local computation of watershed regions can be used as an operator in other segmentation techniques such as seeded region growing, region competition or markers-based watershed segmentation. We illustrate the efficiency and accuracy of the proposed technique on several MRA and CTA data.