In this paper, we propose a post-processing step to enhance the outcome of many corner detectors using the so-called circumferential anchor or CA points. The concept relies on the detection of edges. In our work, we use the SUSAN detector as an example to provide both edges as well as corners, which can be filtered utilizing the edges. Our algorithm places a circular mask at each corner. Points along the circumference that lie on edges are identified as CA points. Radial rays joining each corner and its associated CA points are scanned in the edge domain. According to a strength threshold, a radial ray is accepted. A corner (or a junction) may survive only if at least two radial rays can pass the scanning test. Corners with exactly two rays can be accepted only if those rays do not form a straight line. Experimental results show that this approach reduces the number of erroneous corners along edges and at noisy locations. Also, the results show that corner detectors can be transformed into junction detectors using our algorithm.
展开▼
