Detection of Curvilinear Centerline by Using Hough Voting

Abstract

Extracting centerlines of curvilinear structure in images is a critical step in many computer vision tasks. In this paper, we propose a novel centerline detection approach, which exploits the Hough voting method. We consider the curvilinear structure as a special object, and treat the centerline extraction as an extremum regression problem. Then we construct a voting framework to cast votes for the locations of centerline points based on the generalized Hough transform. The voted local maxima in Hough space constitute the set of potential centerline points with their corresponding scales. In order to obtain the centerlines accurately, we utilize surrounding suppression to counteract the wrong peaks in multi-scale space. The final step is to connect the centerline points to complete the whole centerlines through a simple set of morphological operations. Experimental results show that our method can achieve competitive results compared with the state-of-the-art techniques for various kinds of images.