Unbiased Curvilinear Structure Extraction for Cartoon Images

Abstract

In this paper, a novel method for unbiased curvilinear structure extraction of cartoon images is proposed. Our method based on Delaunay triangulation can extract two types of curves, decorative curves and boundary curves, with sub-pixel accuracy. In order to make it, we start by estimating edge points from both types of curves, using the information of the curves and their surroundings. According to the properties of Delaunay triangulation on edge points, we not only estimate the precise position of curve points, but also find the connection relationships of decorative curve points. Compared with related curvilinear structure extraction algorithm, our method can compute the unbiased position of decorative curve points and boundary curve points and, more importantly, keep the topological relationship of decorative curves well.