January 21 2010
_______________________________________________________________
L0-AbS Deblur Pack -- version 1.0
_______________________________________________________________
Thank you for using this non-commercial research software,
which implements the "L0-AbS (ell-zero Analysis-based Sparsity)
deblurring" method with fixed threshold and combined tight
frames.
It is a Matlab(c) package for non-blind image deblurring (i.e.,
it assumes both the blurring kernel and the noise variance are
known), based on the method described in "IMAGE RESTORATION
THROUGH L0 ANALYSIS-BASED SPARSE OPTIMIZATION IN TIGHT FRAMES",
by Javier Portilla, published in the Proceedings of the IEEE
International Conference on Image Processing (ICIP) in 2009. It
has been tested with the following Matlab versions: R14, 2006b
and 2007b, in Windows XP 32, and 2009a in Linux 64 bits.
This package includes the Matlab(c) open source code files:
Demo shell
==========
(deblurring_L0_AbS_combined_frame_ICIP_2009_demo.m). This code
is intended to be modified easily to run the performance tests
from the ICIP 2009 article cited above, as well as to produce
new degradations (combinations of blurring kernels and noise
variance) and use other images, change the representation used,
the number of iterations, etc.
Simulation function
===================
(simulate_blurry_noisy.m). Called from the previous demo shell,
and it is used for obtaining the degraded used in performance
tests. Note that the blurring is done in the Fourier domain
(which impose its boundary conditions).
Deblurring function
===================
(deblurring_L0_AbS_combined_frame.m)
Besides the above files, the package contains three directories
that should be accessible in the Matlab path:
* Test Images: Some classical test images, including the ones
used in the ICIP 2009 article ("cameraman" and "house").
* Tools: Besides a "psnr.m" function, some useful functions
from publicly available Eero Simoncelli's matlabPyrToolbox
* Pyramids: Include two pyramids. First is the Dual-Tree Complex
Wavelet Transform (DT CWT), from Nick Kingsbury and
collaborators (DT CWT Pack - version 4.3), and two additional
functions to make it compatible with the matlabPyrToolBox (in
the "modified" subdirectory). Second is a spatial
implementation of the Translation Invariant Haar Pyramid
(TIHP), from Javier Portilla, described in detail in J. A.
Guerrero-Colón, L. Mancera, and J. Portilla, “Image restoration
using space-variant Gaussian scale mixture in overcomplete
pyramids,” IEEE Trans. Image Proc., 17(1), 27 – 41, Jan. 2008.
There are two implementation versions of this latter pyramid: a
Matlab(c) MEX C-compiled version (included here as two .dll,
and two .mex64 files) and a open source Matlab(c) version (the
other four .m files). The MEX compiled version is much faster,
but in this release it will only run under Windows 32 or Linux
64 bits.
You may report bugs or other feedback to:
portilla@io.cfmac.csic.es
Enjoy it!
Javier Portilla
Instituto de Optica
Consejo Superior de Investigaciones Científicas (CSIC)
Madrid (Spain, EU).