A MIRROR-CHECKING ALGORITHM FOR EFFICIENT CIRCLE/ARC DETECTION

The iterative multi-step scheme has been applied to circle/arc detection. It usually includes three main steps: picking initial points, finding correspondent searching points with predefined geometric properties, and obtaining candidate circles/arcs. A successful iteration for finding a candidate circle/arc depends on picking valid initial points and finding valid correspondent searching points, and the "valid" means those points must lie on the same target circles/arcs. That is, the iteration is redundant if initial points or searching points are invalid. In this paper, an efficient circle/arc detection method (MCD) based on a mirror-checking algorithm is proposed: we first randomly pick two initial points and construct two corresponding patterns by collecting neighbor points around the initial points. Then, we use the proposed mirror -checking algorithm to check whether the two patterns are mirrored. If they are not mirrored, we will ignore the initial points and find new ones to avoid redundant iteration; otherwise, a candidate circle can be obtained from the mirrored patterns. Instead of finding searching points within a predefined geometric region, the searching point is picked from the mirrored patterns. Based on the initial and searching points, the verification process is utilized to confirm whether the candidate circle is the target circle. From the experimental results, the proposed algorithm can efficiently reduce redundant iterations and executing time. In the meanwhile, we plug the proposed mirror-checking algorithm to the effective voting method (MEVM) and semi-random detection method (MSRD). The experimental results show that it can speed up the two methods and sift redundant iterations efficiently.