Learning multiscale sparse representations: Image restoration and beyond
Abstract for Guillermo Sapiro's Distinguished Lecture, "Learning multiscale sparse representations: Image restoration and beyond"
A framework for learning multiscale sparse representations of color images and video with overcomplete dictionaries is presented in this talk. The dictionary learning is formulated as an optimization problem, efficiently solved by combining quadtree structures with orthogonal matching pursuit (OMP) and one-rank pproximations. The proposed framework provides an alternative to pre-defined dictionaries such as wavelets, and shown to lead to state-of-the-art results in a number of image and video enhancement and restoration applications. We conclude the talk with discussion of learning sparse representations beyond the task of restoration. This talk is based on joint work with J. Mairal and M. Elad. Additional material presented is the result of work with J. Mairal, F. Rodriguez, F. Bach, J. Ponce, and A. Zisserman.