PROJECT TITLE :

Cross-Scale Predictive Dictionaries

ABSTRACT:

An efficient model for signals that do not benefit from analytic sparsifying transformations is provided by sparse representations based on data dictionaries. As a result, sparsifying dictionaries can be computationally intensive for addressing inverse issues, especially when the dictionary under consideration has a significant number of elements In order to speed up the solution of sparse approximation issues, we provide additional structure to dictionary-based sparse representations for visual signals in this study. Sparse models have a multi-scale structure in which each scale's sparse representation is bound by the scale's sparse representation at a lower level. For linear inverse issues connected with photos, movies, and light fields, this cross-scale predictive approach yields large speedups, often in the range of 10-60_, with minimal compromise in accuracy.