Edge linking by a directional potential function (DPF)

Abstract

Low level edge detection operators do not usually guarantee the generation of contiguous boundaries of objects in images. This, coupled with other inherent signal noises, makes many image analysis tasks difficult. We present a new algorithm that is theoretically inspired by a potential function method originated in physics. In this algorithm an edge image is modelled as a potential field with energy depositions at the detected edge positions. Pixels at the edge broken points are charged by the potential forces of the energy in proportion to the relative distances and directions of the surrounding edge pixels. A directional potential function (DPF) is applied to measure the energy charges, which in turn direct the edge connections at these points. Heuristic techniques are utilized in the algorithm to assist the search of charged pixels and improve the effectiveness of the DPF evaluations.