Skip to Main content Skip to Navigation

Détection de Contours et Diffusion Anisotropique dans les Images

Abstract : This thesis is devoted to the anisotropic edge detection and the anisotropic diffusion in images. Based on Gaussian half filters, these half filters scan all orientations around a pixel with a defined step. From oriented derivatives of Gaussian half filters, an approach to edge detection in color images has been developed. Since the filter is long and thin, this method allows to extract anisotropically the contours of the large structures but also of small objects, even in the presence of noise. Gamma curves enhance dark or over-illuminated areas of an image. The combined use of these half filters and gamma curves provide an additional robustness to the method as it is able to extract edges of objects, even in areas under and over-exposed of an image. The second work in this thesis is concerned with the extraction of ridges in the images using the difference of two anisotropic half Gaussians. This new method is very accurate for roof edges even if they are strongly curved or bended. Moreover, the use of half filters provide additional information at level of the junctions which correspond to the local maxima. The final topic of this thesis is dedicated to the anisotropic diffusion in images. Thus, three new regularization schemes have been developed either for removing the textures or for denoising. This last scheme provides interesting results, even in highly noisy color images.
Document type :
Complete list of metadata
Contributor : Baptiste Magnier Connect in order to contact the contributor
Submitted on : Wednesday, January 23, 2013 - 4:06:55 PM
Last modification on : Wednesday, October 20, 2021 - 1:23:01 AM
Long-term archiving on: : Saturday, April 1, 2017 - 8:55:34 AM


  • HAL Id : tel-00780268, version 1



Baptiste Magnier. Détection de Contours et Diffusion Anisotropique dans les Images. Traitement des images [eess.IV]. Université Montpellier II - Sciences et Techniques du Languedoc, 2011. Français. ⟨tel-00780268⟩



Record views


Files downloads