Solving DC programs that promote group 1-sparsity

TitleSolving DC programs that promote group 1-sparsity
Publication TypeConference
Year of Publication2014
AuthorsErnie Esser
Conference NameSIAM Conference on Imaging Science
Month05
Keywordsdifference of convex, group 1-sparsity, nonconvex PDHG, operator splitting, phase unwrapping
Abstract

Many interesting applications require solving nonconvex problems that would be convex if not for a group 1-sparsity constraint. Splitting methods that are effective for convex problems can still work well in this setting. We propose several nonconvex penalties that can be used to promote group 1-sparsity in the framework of difference of convex or primal dual hybrid gradient (PDHG) methods. Applications to nonlocal inpainting, linear unmixing and phase unwrapping are demonstrated.

Notes

(SIAM Conference on Imaging Science)

URLhttps://slim.gatech.edu/Publications/Public/Conferences/SIAM/2014/esser2014SIAMISsdc_pres.pdf
Citation Keyesser2014SIAMISsdc