A potential field-based multi-agent model for contour detection in range images

In this paper, we propose a reactive potential-based multi-agent system for edge detection in range images. A population of situated agents is launched to explore the image. While moving, each agent smoothes the image along its path and alters points which do not belong to surface on which it moves. On the border between two adjacent surfaces, two groups of agents will be in competition to include border points into their respective surfaces. This competition will preserve contours against erasing. However, noise is erased by successive smoothing operated by agents moving on the same surface. A potential field, created around the altered points, allows the agents, moving in the neighborhood, to concentrate their actions around these points. After several alterations of the same point, the potential field around it is relaxed, allowing, agents which were under its influence, to be released and explore other areas of the image.