Analyse statistique de textures directionnelles. Application à la caractérisation de matériaux composites.

Abstract : This thesis aims at the characterization of directional textures i.e. textures composed of oriented structural patterns. More precisely, we show that those textures can be efficiently described through their local orientation field. We tackle three issues.

First of all, we propose a new framework for the conception of operators dedicated to the estimation of an orientation. Two kinds of operators are introduced : the gradient and the valleyness-like operators, which apply respectively on sloped regions and on crest or valley lines. Under the constraint of locality, we present an optimization procedure in order to ensure the precision and the robustness of the approach. Those operators are then combined so as to guarantee a good estimation on any point of the image. Their performances are studied on both synthetic and natural textures.

We then define a statistical method for the description of directional textures, which is based on the second order properties of the local orientation field. More precisely, we use the orientation spatial differences in order to construct orientation-based interaction maps. These maps are applied to the characterization of both natural quasi-periodic textures and composite material images.

Finally, we deal more specifically with the analysis of images obtained from a Transmission Electronic Microscopy technique. These images consist of wave-like directional primitives called fringes. We choose to carry out their characterization by using a structural approach including the identification and the description of the fringes in terms of length and ripple. The ripple phenomenon is studied through the undulation spectrum, that is the spectrum of the orientation computed along the fringes. A stochastic model is finally provided in order to explain the experimental results obtained previously.
Complete list of metadatas

Cited literature [76 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00169937
Contributor : Jean-Pierre da Costa <>
Submitted on : Wednesday, September 5, 2007 - 3:10:28 PM
Last modification on : Monday, February 26, 2018 - 10:02:03 AM
Long-term archiving on : Friday, April 9, 2010 - 1:36:47 AM

Identifiers

  • HAL Id : tel-00169937, version 1

Citation

Jean-Pierre da Costa. Analyse statistique de textures directionnelles. Application à la caractérisation de matériaux composites.. Traitement du signal et de l'image [eess.SP]. Université Sciences et Technologies - Bordeaux I, 2001. Français. ⟨tel-00169937⟩

Share

Metrics

Record views

642

Files downloads

1799