ProjIMA

Implementation of [1] and improvements.

Introduction

Due to the high dynamic of SAR images (because of the strong scatterers), applying a simple total variation (TV) regularization to a SAR image cannot give good results. The aim of [1] is to introduce a TV+L0 decomposition of the scene, in order to have a background regularized with a TV prior and the scatterers with an L0 prior. In this project we apply the TV+L0 decomposition using two methods:

• Ishikawa's graph on the whole image (as suggested in [1] and described in [2]).
• A block-by-block computation of Ishikawa's graph.

The block-by-block computation aims at solving the high-memory required by the construction of the graph described in [2]. Currently, the project implements Rice, Rayleigh and Gaussian distributions to model the noise. It can be used on both SAR (.imw) and "normal" images.

Using the project

This project uses CMake as a build system.

• First, you need to download and install the lib SAR.
• Make sure that CMake is installed on your system.
• In CMakeLists.txt, replace this line to the path to the include dir where lies libSAR.

include_directories(/Path/to/libSAR/include/)

• Go in the directory of the project and type:

cmake . && make

• You should have two binaries created in bin/. One of them (Viewer) is a very simple viewer for imw images. The other one, named Denoise works that way:

./bin/Denoise -BBV 1 -BS 30 -i inputName -oBV outputNameBV -oS outputNameS -oC outputNameC [-r]

This line will apply the decomposition on an image named inputName and will produce 3 images:

• outputNameBV (The background image);
• outputNameS (The scatterers image);
• outputNameC (The complete image, i.e. + outputNameBV + outputNameS).

If you are using a radar image (e.g. image.imw bundled with image.dim), just put image (without the extension) as inputName and add the -r option.

Other options are available and can be checked by calling Denoise with -h. One of them is -no which will use the block-by-block computation.

Bibliography

[1] L. Denis, F. Tupin and X. Rondeau, "Exact discrete minimization for TV+L0 image decomposition models", ICIP 2010, Hong Kong, September 2010.

[2] H. Ishikawa, “Exact optimization for Markov random fields with convex priors,” IEEE TPAMI, vol. 25, no. 10, pp. 1333–1336, 2003.