Hierarchical Geometrical Image Representation with Plain Wavelets

Stéphane Mallat
Ecole Polytechnique, Palaiseau France/CMAP

Is it possible to build geometric image representations with local transformations over separable wavelet coefficients ? Following this approach, we construct wavelet-bandelet orthogonal bases, that lead to asymptotically optimal approximations of geometrically regular images. These bases remain orthogonal for discrete images and their structures share some similarities with neurons in the V2 visual area.

Back to MGA Workshop I: Multiscale Geometry in Image Processing and Coding