Active contours using a potential field

Abstract

In most implementations of active contours (snakes), the evolution of the snake depends only on image characteristics in the immediate neighborhood of the current snake points. This is true even when there is little edge data available in the current neighborhood, and even when the boundary of interest may be some distance away in the image. This paper proposes a vector potential field at each point in the image that is derived from the "pull" exerted by all edge points in the image; the pull for a given edge is inversely proportional to the square of the distance from the pixel it pulls. This potential field acts as a force, and snake points are moved based on the force at their current location, rather than moving to minimize energy at a candidate position. The resulting algorithm allows edges to influence snake evolution earlier and from a greater distance, and results in faster and better convergence to the final boundary under a variety of image characteristics.