Contour segmentation using Hough transform

In many 2D situations, contours can be efficiently represented as line segments or short circle arcs sequences. Therefore, we propose an original use of the Hough transform for detecting such primitives. Rather than searching directly for circle arcs, we have chosen to use the Hough transform for evaluating the relevance of each edge point and the curvature center associated with it. A specific parametrisation, which takes into account the discrete aspect of the problem, provides an homogeneous Hough space and bounds its spatial complexity. Then, circle arcs are obtained by gathering points which are close and have equivalent curvatures. An application to medical imaging is presented to highlight the significance of this method.