Joint Ellipse and Line Detection Using A Contrario Theory

Viorica Pătrăucean, Rafael Grompone von Gioi, Pierre Gurdjos

Online Demo

Source Code

The source code of ELSDc is available here (C language).

Abstract

Given as input a grayscale image, we seek to summarise its content in terms of line segments and elliptical arcs. The main concern in designing geometric feature detectors is to obtain reliable results (i.e. reduced number of false detections) on various types of images without any parameter tuning. To achieve this goal, we propose an automatic procedure, having a twofold decision-making role: model validation and model selection. For a given set of pixels in a grayscale image, the detector decides if a feature is present or not (model validation), and identifies its type when multiple interpretations are possible (model selection). We describe a continuous parameterless criterion based on the a contrario theory which serves as validation and model selection criterion. Altought we are interested in detecting line segments and elliptical arcs, the elementary geometric features considered in our study are polygonal chains and elliptical arcs, which ensures a fair model selection. The experimental results show that our method achieves superior precision compared to state-of-the-art detectors when applied on synthetic or real images without any parameter tuning.