deconvlab / sbd-rtrm

Sparse blind deconvolution using Riemannian Trust Region Method.

SBD-RTRM: Sparse blind deconvolution using the Riemannian Trust-Region Method (RTRM)

SBD-RTRM is a MATLAB package for sparse blind deconvolution (SBD) using the Riemannian Trust Region method (RTRM). As sparse blind deconvolution is a nonconvex problem, using RTRM ensures that local minima will be found in the associated optimization objective.

Our package is motivated by studies in blind deconvolution as a nonconvex optimization problem, and by applications in Scanning Tunneling Microscopy.

For documentations, info and references see docs/README.ipynb.

Updates

2018-03-16: -Implemented backwards Compatibility with Xsolve_pdNCG. -Multiple slices now works correctly for Xsolve_FISTA.

2018-02-19:

• Removed signflip option
• Option getbias to estimate constant bias is added across SBD.m, Asolve_Manopt.m and Xsolve_FISTA.m. The pdNCG solver is now depreciated in terms of both Xpos and getbias.

2018-01-31:

• Option to solve for X>=0 Xpos is included.
• Xsolver changed from pdNCG to FISTA, and the sparsity surrogate is changed from pseudo-Huber to Huber function.

Upcoming changes

• Adding a reweighting method to sharpen recovered activation maps
• Account for border effects