A quantitative comparison of state of the art circle detection algorithms

Detecting circular objects in digital images are crucial problem in common applications. Although several circle detection algorithms have been released in the literature, the algorithms utilize a small set of images to show effectiveness. This situation causes unfair comparison between algorithms. In this thesis, a dataset including 200 images with size 800x6000 and and human annotations are proposed. Images in dataset have circular objects chosen from several application areas. The collected dataset is named as Anadolu University Circle Detection Dataset and Benchmark (AUCDB200), and is carried out for quantitatively comparison of the state of art circle detection algorithms in precision-recall-Fscore metrics. In this thesis, a novel circle detection algorithm is also proposed with benefiting from circular arcs of recently proposed Orientation Transform (OT). The novel algorithm is named as OTCircles. The experimental results in the thesis show that proposed algorithm, OTCircles, presents the best performance for proposed AUCDB200 dataset with 0.92 Fscore. The another results demonstrates that the algorithm is more robust against to noise.