Multi-Resolution algorithms for Active Contour Models

Abstract : Multi-resolution methods applied to active contour models can speed up processes and improve results. In order to estimate those improvements, we describe and compare in this paper two models using such algorithms. First we propose a multi-resolution algorithm of an improved snake model, the balloon model. Convergence is achieved on an image pyramid and parameters are automatically modified so that, at each scale, the maximal length of the curve is proportional to the image size. This algorithm leads to an important saving in computational time without decreasing the accuracy of the result at the full scale. Then we present a multi-resolution parametrically deformable model using Fourier descriptors in which the curve is first described by a single harmonic; then harmonics of higher frequencies are used so that precision increases with the resolution. We show that boundary finding using this multi-resolution algorithm leads to more stability. These models illustrate two different ways of using multi-resolution methods: the first one uses multi-resolution data, the second one applies multi-resolution to the model itself.
Type de document :
Communication dans un congrès
Springer Berlin Heidelberg. International Conference on Analysis and Optimisation of Systems, Images, Wavelets and PDE's (ICAOS), Jun 1996, Paris, France. 219, pp.58-65, 1996, Lecture Notes in Control and Information Sciences. 〈10.1007/3-540-76076-8_117〉
Liste complète des métadonnées

https://hal.inria.fr/inria-00532688
Contributeur : Brigitte Briot <>
Soumis le : jeudi 4 novembre 2010 - 13:38:50
Dernière modification le : vendredi 25 mai 2018 - 12:02:05

Lien texte intégral

Identifiants

Collections

Citation

Bertrand Leroy, Isabelle Herlin, Laurent Cohen. Multi-Resolution algorithms for Active Contour Models. Springer Berlin Heidelberg. International Conference on Analysis and Optimisation of Systems, Images, Wavelets and PDE's (ICAOS), Jun 1996, Paris, France. 219, pp.58-65, 1996, Lecture Notes in Control and Information Sciences. 〈10.1007/3-540-76076-8_117〉. 〈inria-00532688〉

Partager

Métriques

Consultations de la notice

310