MIT Department of Electrical Engineering & Computer Science

MIT EECS
1994 (Fall Semester)
Colloquium Series

Monday, November 7, 1994

ADAPTIVE SIGNAL REPRESENTATIONS WITH WAVELETS

Stephane Mallat
Courant Institute

The wavelet transform is used to approximate and analyze time or space varying structures. A first application to the characterization of singularities and multiscale edges is presented. The diversity of transients in complex signals requires to adapt the wavelet type to the signal components. A matching pursuit algorithm performs such an adaptive expansion by choosing the wavelet among a large and redundant dictionary. With a dictionary of Gabor functions, it yields an adaptive time-frequency transform which is used for music processing. Images are decomposed over dictionaries of oriented and dilated wavelets for edge and texture analysis.

November 7, 1994
4-5pm
Grier Room (34-401)
Refrehment at 3:30 pm