University of California Irvine
Harvesting sparsity in imaging sciences
Sparse representation has attracted tremendous attention over the past decade. Particularly in imaging sciences, it has been widely employed as prior information to facilitate a wide spectrum of image reconstruction problems, where the underlying image can be sparsely represented under a set of basis functions. One important issue behind these problems is the choice of such basis functions. An appropriately chosen basis set can have better sparse representation of an image, thus yielding better results. In this talk, we target the basis selection in the problem of image deblurring. Our proposed algorithm outperforms the state-of-the-art, especially in deblurring text images. During the course, we have observed a limitation of this approach in terms of coherence. It is the same limitation that many conventional methods have when solving sparsity-related problems. We then propose a novel technique that can promote sparsity even in highly coherent scenarios. Satisfactory experimental results have been observed; and theoretical aspects behind this approach have also been discovered.
Sponsored by the Department of Mathematical Sciences
Host: Sue Minkoff
Refreshments will be served in Room TBD 30 minutes before the talk begins.