Extraction d'éléments curvilignes guidée par des mécanismes attentionnels pour des images de télédétection. Approche par fusion de données

Authors: Cotteret, Gilles
Advisor: Edwards, Geoffrey; Ligozat, Gérard
Abstract: The extraction of curvilinear elements from remote sensing images, especially when noisy or near the limit of resolution constitutes a significant challenge for data-processing algorithms. In this work a method is presented for linear feature extraction in remote sensing (RS) images. An original model (ELECA) is introduced allowing out of date geographical information system (GIS) data to be updated though the use of a visual search method that mimics human eye movements. The ELECA model is composed of three parts : (1) a visual search module using virtual gaze to avoid processing the entire image ; (2) a simple and fast method for local information extraction by a clever adaptation of connected-component labeling ; and (3) an original method for the fusion of local information to construct a global representation at the scale of the image based on qualitative spatial reasoning techniques. The ELECA model avoids several problems characteristic of current methods. In particular, the proposed technique can be applied to low resolution or partially occluded images for which currently only human interpreters can successfully process the image. The technique is also designed to be very fast and efficient when a quick GIS update is needed. The last part of this project is devoted to the design of software which supports the ELECA model. The proposed software architecture is adaptive and allows the integration of future model developments. Finally it is shown how the ELECA model could be implemented.
Document Type: Thèse de doctorat
Issue Date: 2006
Open Access Date: 11 April 2018
Permalink: http://hdl.handle.net/20.500.11794/18240
Grantor: Université Laval
Collection:Thèses et mémoires

Files in this item:
23332.pdf9.71 MBAdobe PDFView/Open
All documents in CorpusUL are protected by Copyright Act of Canada.