Fast circle detector based on region-growing of gradient and histogram of euclidean distance

The most existing circle detectors based on Hough transform need to tune many parameters while the methods based on histogram are complex in computation and resource, thus a fast circle detector based on regiongrowing of gradient and histogram of Euclidean distance is presented to solve the above problems. The pixels' gradient module and direction are computed in the first step and region-growing method is implemented to generate arc support regions. Three coordinates of each arc support region (ASR) are then selected to solve the center and radius of its corresponding circle and determine a square fitting area (SFA). Afterward, the Euclidean distances between every coordinates on each ASR and each coordinate of its ASR's corresponding SFA are computed and recorded in a three dimensional accumulator. A histogram is used to count the frequency of the distances that participate in the accumulator and the parameters of each circle are acquired. A verification strategy of circular integrity is used to test the detection results. Compared with the histogram based circle detection (HBCD) and random Hough transform (RHT), experimental results indicate that the proposed algorithm is able to detect partial circles, multiple centers or circles in partial occlusion. This method has features of high speed, low consumption, wide range of application and strong anti-interference performance.