Watersnake: integrating the watershed and the active contour algorithms

Abstract

We propose a new segmentation technique by integrating the active contour (snake) and the watershed algorithms. The two distinct techniques for image segmentation are combined in a manner which utilizes the strengths of both. Objects obtained by the watershed transform lack the smoothness of those obtained by contours. The accuracy of active contours, on the other hand, depends upon contour initialization around each object of interest. A new segmentation technique is proposed to eliminate the problems encountered by each method. The watershed-based segmentation on a down-sampled image is used to get the initial contours for the active contour models. The proposed integration incorporates control of smoothness in watershed, makes active contours unsupervised and prevents the snake from getting trapped in local minima.