A novel topology based watershed segmentation

We propose a novel method for watershed segmentation based on the topological properties of triangular irregular networks (TIN) associated with input images through their height fields. The classical watershed segmentation is very sensitive to the initial markers definition. In order to avoid undesirable effects such as oversegmentation, we propose to use topological features, such as critical points, to extract meaningful markers. Critical points based watershed algorithm is developed and implemented to carry out grey scale image segmentation. Unlike the classical watershed algorithms, which rely on a flooding simulation starting from the gradient´s image minima, the proposed technique defines growing regions for both maxima and minima. It therefore uses the grey scale image directly and avoids noise amplification that results from the gradient operator. Experiments demonstrate that this method provides a good segmentation procedure for gray-scale images.