Finding the N-cuts of Watershed Partitions for Image Segmentation

The normalized cut (N-cut) algorithm uses an algebraic graph optimization technique for image segmentation. Although N-cut produces good results for a variety of images, it has some weaknesses, such as high computational cost and sub-optimal partitions. In this paper we adopt the watershed transform to address these problems. Watershed can improve slow computing speed and produce closed object boundaries. However, watershed itself has the drawback of over-segmentation. Therefore, we propose to first apply watershed, then build a graph from the watershed regions, and find the N-cuts of the watershed region graph to improve segmentation accuracy. The objective of this paper is two-fold; the first goal is to reduce the complexity of this problem by optimizing region-based graph structures. The second goal is to validate the performance of the existing and proposed methods, and to test the hypothesis that region-based analysis reduces the complexity of optimization problem and improves segmentation accuracy.