Effective arc extraction method for ellipse detection in complicated images

Due to corruption from image clutter and noise in Complicated natural images, ellipses are likely to be broken into many small disconnected arcs which is challenging for ellipse detection methods based on edge following techniques. In order to cope with this problem, a novel arc extraction method is proposed based on edge curvature. The method first groups connected pixels to form line segments and then breaks edge contours into smooth arcs at the turning points, inflexion points and noise connection points of the curve. Finally, near distance arcs are reconnected based on edge curvature. In this aspect, two arcs are linked if the new arc has no turning points, inflexion points and noise connection points. Additionally, we apply the arc reconnection method on a set of challenging images and the ellipse detection results get improved. Therefore, this work contributes towards ellipse detection.