Research On Circle Recognition Under Complex Background

Circular recognition has been a hot issue in the field of computer vision for several years. What we expound in this paper is how to recognition the circular targets in complex background. The Hough transform-based approach is widely used in solving this problem. Because of the inefficiency of traditional circular Hough transform, a lot of improved algorithms were proposed. In this paper, we propose a algorithm which is based on improved circular Hough transform and support vector data description to complete circular recognition.The main contents of the paper include preprocessing, center detection, radius detection and false circle recognition. The original image we obtained from hardware devices may not meet the requirements. Preprocessing was used to separate the target from the background. Center detection was used to detect the candidate centers. Radius detection was used to find out the group of optimal radius. False circle recognition was used to delete the false circles from the candidate circles.It is necessary to do a series of preprocessing to improve image quality, highlight the target and reducing background information, because the input images were in complex background. Meanwhile, edge detection in preprocessing lead us to do circular recognition in edge image rather than in original image which is of complex pixel information. Circular recognition in this paper is divided into center detection and radius detection. Center detection is completed via improved Hough transform, image segmentation and clustering analysis. A SVDD-based algorithm is used to complete radius detection, we do false circle recognition according to some of the parameters of the SVDD-based algorithm. Then we finish the whole work of circular recognition. Three sets of comparative experiments were finished to test the effectiveness and replicability of our algorithm.The main innovations of this paper include:(1) Improving the Hough transform;(2) The use of SVDD in proposing optimization algorithm to determine the radius of the circles.