Linear band detection based on the Euclidean distance transform and a new line segment extraction method

Abstract

A linear band, which is a straight line segment with some width (i.e., thickness), is a more structured, higher-level feature compared to edge or line features. In spite of the usefulness of linear bands as features, papers dealing with their detection problem are rare. In this paper, we propose a new method for detecting linear bands in gray-scale images. We first talk about our opinion on what types of linear bands a desirable detector should be able to detect, and then give a description on how we designed our detector to achieve the goal. Our method consists largely of two parts: (1) extracting the candidate center line pixels of the linear bands contained in an input gray-scale image (sub-parts: edge detection, Euclidean distance transform, ridge detection in a distance map, and noisy ridge pixel removal), (2) extracting line segments from the result of (1) using our new line segment detection method (sub-parts: modified Hough transform, base line segment grouping, redundant line segment removal, and postprocessing). Experimental results show that our method is practical and robust.