Testing for Curves in a Binary Image

Curve detection is viewed as a process of hypothesis gener- ation and hypothesis testing. Of the two, hypothesis generation has re- ceived much attention and many sophisticated post-processing strategies are published in the literature. In this work, the emphasis is shifted to the development of an efficient and effective hypothesis testing strategy to relieve the hypothesis generation from sophisticated computations. The proposed method recasts edge pixels in a binary image into a one- parameter system derived from the hypothesis. The recasting process creates a histogram, which contains a single dominant peak if and only if the hypothesis under testing contains a significant number of edge pixels. Experiments with circle testing show that the proposed strategy outperforms the global threshold.