Conférencier invité:
Wim Sweldens
Bell Laboratories, Lucent Technologies
http://wim.sweldens.com
Destinés aux étudiants, aux post-doc et aux chercheurs (du milieu académique ou non), ces ateliers sont composés de deux parties: un mini-cours en avant-midi (d'une durée de deux heures environ) et un séminaire en après-midi, plus spécifique aux travaux du conférencier. L'atelier du 16 février porte sur les plus récentes avancées dans le domaine de la théorie de l'approximation par des techniques multi-échelles. Le conférencier invité est l'expert en la matière: en selectionnant le Dr. Wim Sweldens pour faire partie du dernier TR100 (sélection des 100 meilleurs jeunes chercheurs américains), le comité soulignait la qualité de ses contributions à la théorie des ondelettes ainsi que le haut degré d'innovations que ses travaux engendraient dans les domaines de la communication, de la synthèse d'images et du traitement du signal.
Prochain atelier: Data Minnig (Prof. H. Chipman, Univ. of Waterloo), première semaine d'avril 2000.
MINI-COURS
Le mercredi, 16 février 2000, 10h30 CRM, Université de Montréal Pavillon André-Aisenstadt 2920, Chemin de la Tour, salle 5340
"The lifting scheme and second generation wavelets"
In the last decade wavelets have been applied successfully to sound (1D), image (2D), and video (3D) processing. Typical applications include compression, noise reduction, progressive transmission, etc. Each time the data is defined on an Euclidean space and sampled on a regular grid. Many applications, however, need wavelets defined on general geometries (curves, surfaces, manifolds), wavelets adjusted to irregular sampling, or adaptive wavelet transforms. Therefore we introduce Second Generation Wavelets: wavelets which are not necessarily translates and dilates of one function, but still enjoy all powerful properties such as time-frequency localization, multiresolution, and fast algorithms. While the Fourier transform has been the principal tool in constructing classical wavelets, e.g. Daubechies, it can no longer be used to build Second Generation Wavelets. We therefore present the lifting scheme, an entirely spatial construction technique.
We give examples how lifting can be used to build wavelets for irregular samples, spherical wavelets, and multiresolution geometry. We also show that all classical wavelets can be obtained through lifting, that lifting speeds up the fast wavelet transform, and that lifting allows for integer-to-integer wavelet transforms which are important in lossy compression.
Note: No preliminary knowledge of wavelets will be assumed.
SÉMINAIRE
Le mercredi, 16 février 2000, 15h30 CRM, Université de Montréal Pavillon André-Aisenstadt 2920, chemin de la Tour, salle 5340
"Digital Geometry Processing"
Due to the advancement in 3D scanning techniques it now is relatively easy to build accurate digital descriptions of complex geometrical objects and scenes. Digital geometry has found numerous applications in telecommunications, medicine, and entertainment. This has set off "digital geometry processing" as a new branch of digital signal processing. However, many of the standard digital signal processing tools are not suited for digital geometry. In this talk I will show how recently developed multiresolution and wavelet techniques can address digital geometry tasks such as compression, storage, transmission, editing and animation.
Information: Jean-Marc Lina (lina@CRM.UMontreal.CA) 343-6111, poste 4746