A Contrario Line Segment Detection

verfasst von: Rafael Grompone von Gioi

Verlag: Springer New York

Buchreihe : SpringerBriefs in Computer Science

Enthalten in: Springer Professional "Wirtschaft+Technik" , Springer Professional "Technik" , Springer Professional "Wirtschaft"

Einloggen, um Zugang zu erhalten

Über dieses Buch

The reliable detection of low-level image structures is an old and still challenging problem in computer vision. This book leads a detailed tour through the LSD algorithm, a line segment detector designed to be fully automatic. Based on the a contrario framework, the algorithm works efficiently without the need of any parameter tuning. The design criteria are thoroughly explained and the algorithm's good and bad results are illustrated on real and synthetic images. The issues involved, as well as the strategies used, are common to many geometrical structure detection problems and some possible extensions are discussed.

Inhaltsverzeichnis

Frontmatter

Chapter 1. Introduction

Abstract

The aim of image analysis is to extract useful information from digital images. The fundamental problem lies in how to combine the information from multiple measurements, the pixel values, into a coherent interpretation. This work discusses a very reduced version of this general problem, namely the detection of line segments.

Rafael Grompone von Gioi

Chapter 2. A Contrario Detection

Abstract

At the turn of the century, Desolneux, Moisan, and Morel undertook the task of formalizing the Gestalt theory into a mathematical framework [19, 22]. The motivation for this ambitious project was to provide a foundation for computer vision based, like the Gestalt theory [46, 51, 61, 62], on a small set of fundamental principles. They identified the lack of a principle to guide the selection of detection thresholds and their main contribution was to propose the a contrario framework to cover this need. This chapter will introduce the a contrario approach and its application to line segment detection.

Rafael Grompone von Gioi

Chapter 3. The LSD Algorithm

Abstract

This chapter describes in full detail the LSD algorithm [31, 35, 36] for line segment detection. It is based on the a contrario framework described in the previous chapter, but instead of searching exhaustively for line segments, it uses the heuristic search plus validation approach, resulting in an efficient algorithm. The source code and an online demo for LSD are available at [36].

Rafael Grompone von Gioi

Chapter 4. Experiments

Abstract

This chapter presents some experiments to illustrate the behavior of the a contrario line segment detector, indicating the good properties as well as its shortcomings. The results are compared with some existing approaches, concluding with an empirical evaluation of the algorithm computational time.

Rafael Grompone von Gioi

Chapter 5. Extensions

Abstract

No existing algorithm for line segment detection can be said to perform this task to full satisfaction. The problem is indeed ill-posed: people often disagree about the exact set of line segments that corresponds to a particular image, revealing differences in the basic notion or in the interpretation of a given scene. But current algorithms are still far from this level of disagreement, producing failures universally recognized as such. As a simple example, none of the full scale results in Fig. 4.17 is satisfactory. These failures essentially show that the definitions of line segment implicitly used are not good enough.

Rafael Grompone von Gioi

Backmatter

Titel: A Contrario Line Segment Detection
verfasst von: Rafael Grompone von Gioi
Verlag: Springer New York
Electronic ISBN: 978-1-4939-0575-1
Print ISBN: 978-1-4939-0574-4
DOI: https://doi.org/10.1007/978-1-4939-0575-1