Blind Deconvolution using Convex Programming

This page provides software to generate the figures and the experiments in the paper Blind Deconvolution using Convex Programming. We also provide software created by other groups which is necessary to run our own code.

Required Toolboxes

The following toolboxes are required to run the MATLAB scripts below. The paths to the associated directories need to be provided in the script.

• minFunc
• minFunc_2012
• Noiselet Toolbox

Matlab scripts

We provide the Matlab scripts that generate the figures, as well as a test file that demonstrates large scale blind deconvolution using convex programming.

• Script for blind deconvolution: test.m

  • Requires all three toolboxes
  • Requires: blindDeconvolve_implicit.m

• Script to generate Figure 3 (phase transitions): generateFig3.m

  • Requires minFunc and minFunc(2012)
  • Requires: blindDeconvolve.m

• Script to generate Figure 4 (phase transitions): generateFig4.m

  • Requires minFunc and minFunc(2012)
  • Requires: blindDeconvolve.m

• Script to generate Figure 5 (recovery in the presence of noise): generateFig5a.m and generateFig5b.m

  • Requires minFunc and minFunc(2012)
  • Requires: blindDeconvolve_implicit.m

• Script to generate Figures 6, 7, and 8 (image deblurring): deblur.m

  • Requires minFunc and minFunc(2012)
  • Requires: blindDeconvolve_implicit_2D.m
  • Image data: shapes.png

References

The Noiselet toolbox is by Professor Romberg, while the minFunc toolbox is taken from the following paper:

• M. Schmidt, "minFunc: unconstrained differentiable multivariate optimization in Matlab", 2012.

If you use either of these files in your personal work, please remember to cite these references.