James Nagy, Emory University
Keywords: Iteratively Reweighted Deconvolution through Subspace Projection
Abstract:
Iteratively reweighted deconvolution algorithms can be used to incorporate sparse constraints and to account for outliers in measured data, such as glints. The reweighting process requires solving a sequence of deconvolution problems with different weighted convolution operators. This can be expensive, especially in the case of spatially varying blurs. In this work we describe a general framework to efficiently solve the sequence of deconvolution problems using Krylov subspace projection methods (i.e., conjugate gradient type methods). The projection approach allows much of the difficult work to occur on low dimensional subspaces, and thus significantly reduce the computational cost for large-scale problems.
Date of Conference: September 15-18, 2015
Track: Poster