Circle Detection By Midpoint Circle Validation

Automatic circle detection is the actual demands of real life and industrial production. At the same time, it is a vital problem in image processing. It has been applied to many fields, such as circular traffic sign detection, pupil and iris detection, aiding vectorization of line drawing images, cells in the analysis of medicine, product size detection, sophisticated quality detection, agricultural machinery automation, and so on. Circle detection in digital images is commonly performed by the Circular Hough Transform, the Randomized Hough Transform, the Randomized Circle Detection, and recent proposed optimization-based circle detection. The disadvantage of these methods is that the speed is slow in multiple circle detection, the residual accuracy of circle radius and center is not high, the efficiency of processing complex image is low, and it still has a bad robustness. We lever Midpoint Circle Algorithm to improve the speed of multiple circle detection and the residual accuracy of circle radius and center.The first step is to generate the binary image by Canny operator. Because Canny operator has a better edge connection degree and a fine edge planning than other edge operators. At the same time, it maintains a good anti-noise performance and high positioning accuracy of the edge. Then, we take the circle which is computed with three non-collinear edge pixels in edge map as the possible circle. The third step is to validate the true circle by Midpoint Circle Algorithm (MCA). Because MCA avoids computing square-root calculations by comparing pixel separation distances. It is also considered the quickest providing a sub-pixel precision. At last, since three pixels are randomly picked, the circle parameters are computed by the least square with the edge pixels which close to the true circle. Experimental results validate that the proposed technique not only correctly detects circles, but also has a better speed in multiple circle detection and a lower residual of circle radius and center.