Eliminating False Positives of Hough Transform with Constructive Testing in Line Detection

Abstract

Routines for Hough transform for line detection use multiple parameters that can be difficult to determine sometimes. Even when a good deal of effort is spent on determining these pa-rameters, the true positives found are often accompanied by some false positives. In this paper, we propose a method called constructive testing for post-processing the lines returned by Hough transform routines with the objective to eliminate false positives. Given a detected line, constructive testing builds a small set of parallel lines based on it. Then, the detected line's distinctiveness as a line in contrast to the other constructed par-allel lines is computed with sample statistics. The detected lines are accepted or rejected based on their distinctiveness. Experimental results show determining a threshold of distinc-tiveness is intuitive and easy and that it effectively eliminates a large number of false positives.