ACCV 2016 Tutorial on Digital Geometry Processing
Abstract
Extracting high-quality and robust geometric informations plays an important role many computer vision applications. For instance, the noise, scale or curvature estimations can influence the resulting quality of shape recognition, matching or indexation. In this context, the aim of this tutorial is to present the new robust geometric estimators (like normals, tangents, curvature and local noise estimators), recently proposed in the digital geometry community, starting from their theoretical description up to their concrete implementation in the emerging DGtal Library framework, and to show some of their applications. Such tutorial will benefit to the audience by showing them how to create image analysis tools using these features and how to integrate them in another framework like OpenCV. Also, the audience will exploit the described advanced estimators and discover progressively the DGtal Library through several concrete tutorial applications, starting from simple topology based contour extraction to unsupervised noise detection techniques. The last part of this tutorial will focus on how to build online demonstration tools. This is especially important in image filtering and analysis, in order to make unbiased and reproducible research. This tutorial will present the new automated demonstration framework of the IPOL journal (Image Processing On-Line).
Time and location
20th at 201D room 9:00am-12:20pmLecturer
Bertrand Kerautret
PhD thesis in Computer Science, (Assistant Professor)
LORIA, University of Lorraine
email: bertrand.kerautret@loria.fr
Bertrand Kerautret received a Ph.D degree in Computer Science from Bordeaux I University (France) in 2004. He is currently Assistant Professor in Computer Science at Lorraine University (IUT of Saint DiƩ des Vosges, France) and works in the LORIA laboratory. His research interests are related to computer vision problems, such as 3D shape reconstruction, image analysis and discrete geometry. He also works on discrete geometric estimators adapted to noisy discrete objects. He is interested too in their applications in computer vision, and also in archaeological, medical and agronomic imaging research. He has written about forty papers in international journals or conferences on these topics. He is one of the main contributors of the DGtal Library and the manager of DGtalTools, the part of DGtal dedicated to applications. He is also an active member of the IPOL Editorial Board.
The program schedule
First Part 9h00 - 10:30
- Short Overview of Digital Geometry
- Geometric Estimation on Digital Contours
- Main Primitives used in Digital Geometry
- Estimating curvature on (noisy) digital contours
- Application to contour representation
- Presentation of the DGtal Library
- Handling the DGtal Library to extract level sets contours (with exercice)
- Estimating Geometric features (with exercice)
Cofee Break (10h30 - 11h00)
Second Part 11h00 - 12:30
- Meaningful Scale and Noise Detection
- Presentation of the method
- Implementation in the DGtal Framework
- Highlighting Reproducible Research
- Valuation of implementation in RR journal (IPOL)
- Preview of the new online demonstration (demo generation)