Generalized Hough Transform: Fast Randomized Multi-Circle Detection

To determine a candidate circle,we randomly pick two sample edge points and search the third edge point on their perpendicular bisector.When randomly sampling two points,invalid sample is decreased by recognizing isolated noises and half-link noises.Above two kinds of noises and non-co-circle points are eliminated in the course of searching third point of the possible circle,and a measure is presented to affirm quickly the possible circle for true circle so that invalid computation is decreased.The experimental results demonstrate that the approach can detect quickly multiple circles.The detection speed is an order of magnitude faster than Randomized Circle Detection （RCD） proposed by Chen 牔 Chung （2001） on the condition of detecting multiple circles with noises.