Méthodes robustes en vision : application aux appariements visuels

Zhong-Dan Lan 1
1 MOVI - Modeling, localization, recognition and interpretation in computer vision
GRAVIR - IMAG - Graphisme, Vision et Robotique, Inria Grenoble - Rhône-Alpes, CNRS - Centre National de la Recherche Scientifique : FR71
Abstract : This thesis is devoted to the visual correspondence problem. The robustness and precision problems are considered. The partial occlusion problem is treated using robust statistics. Two methods are introduced, one based on robust regression and the other based on robust location. Experiments validate the methods. Precise matching is also considered. Based on local linearity assumption, local translation is estimated using a convolution mask between two image windows under translation. In non translational case, we estimate robustly the image windows deformation, followed by the estimation of local translation on the deformed window. Experiments validate the method. Matching, being a major issue in vision, is still far away to be solved. We hope this report gives some contributions to this difficult problem.
Document type :
Theses
Complete list of metadatas

Cited literature [100 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00004950
Contributor : Thèses Imag <>
Submitted on : Friday, February 20, 2004 - 5:07:50 PM
Last modification on : Wednesday, April 11, 2018 - 1:52:16 AM
Long-term archiving on: Friday, April 2, 2010 - 8:02:07 PM

Identifiers

  • HAL Id : tel-00004950, version 1

Collections

INRIA | IMAG | CNRS | UGA

Citation

Zhong-Dan Lan. Méthodes robustes en vision : application aux appariements visuels. Interface homme-machine [cs.HC]. Institut National Polytechnique de Grenoble - INPG, 1997. Français. ⟨tel-00004950⟩

Share

Metrics

Record views

398

Files downloads

564