A FAST RANDOMIZED METHOD FOR EFFICIENT CIRCLE/ARC DETECTION

Circle/arc detection plays an important role in image processing and machine vision. The Hough transform has been applied to circle/arc detection, and many multi-step based methods have been proposed for improving its performance (computation and storage space). The multi-step iterative procedure to find candidate circles/arc includes: picking initial points, finding correspondent searching points with some predefined geometric properties, and obtaining candidate circles/arcs. However, the number and distribution of the initial points are keys for efficient detection. In this paper, we propose a new circle/arc detection method, Fast Randomized method for Efficient Circle/arc Detection (FRECD). It just requires one "neighbor" point of target circles/arcs as the initial point. Besides, the proposed FRECD does not use storage for voting space. From the experimental results, the proposed FRECD provides better performance than previous multi-step based methods.