Sparse image fusion

This document is structured as follows:

Sparse image fusion

Description and literature

This software suite comprises the [ SparseFI - J-SparseFI - J-SparseFI-HM ] family of multi-sensor image fusion algorithms for multi- and hyperspectral image super-resolution based on sparse representations.

A detailed description of the algotihms implemented in this software suite is provided in my dissertation:

C. Grohnfeldt, "Multi-sensor Data Fusion for Multi- and Hyperspectral Resolution Enhancement Based on Sparse Representations ", Ph.D. Dissertation, Technical University of Munich, 2017; doi:10.14459/2017md1366768

Getting started

Prerequisites

This code has been tested on UNIX-based machines only. Its dependencies are as follows: gcc/g++, OpenMPI, curl, wget, grep.

Installing third party libraries

This software suite depends on two external libraries, which need to be installed and linked to the repository's main directory.

It is recommended to install those libraries outside of the repo and link them via symbolic links, as instructed further below.

Create a directory, somewhere outside of the repository, into which third party libraries will be installed.

mkdir <chosen-path-to-thirdparty-library-dir>

Eigen: C++ template library for linear algebra

The Eigen project is hosted on http://eigen.tuxfamily.org. A git mirrow is available on GitHub. We'll clone that into the above-created lib directory as follows:

git clone https://github.com/eigenteam/eigen-git-mirror.git <chosen-path-to-thirdparty-library-dir>/eigen

That's it for Eigen. Further below, we will link this library to our repository's directory.

GDAL: Geospatial Data Abstraction Library

GDAL can be installed either on a system level or locally following the descriptions on www.gdal.org. Here, we'll build it from source code and install it locally to avoid dependency on sudo permissions.

Download the source code of the latest stable release:

cd <chosen-path-to-thirdparty-library-dir>
mkdir downloads
cd downloads
tmp=$(curl http://download.osgeo.org/gdal/CURRENT/ | grep -o "gdal-[2-9].[0-9].[0-9].tar.gz")
filename=${tmp/.tar.gz*/.tar.gz}
wget "http://download.osgeo.org/gdal/CURRENT/$filename"
tar zxf $filename

Install locally as follows:

cd ${filename/.tar.gz/}
gdal_prefix="<chosen-path-to-thirdparty-library-dir>/gdal"
./configure --prefix=$gdal_prefix
# for compilation on the SuperMUC, you need to load a gcc
# compiler that is more recent that the default one:
# module load gcc/8
make
make install

Test if local installation process was successfull:

export PATH=${gdal_prefix}/bin:$PATH
export LD_LIBRARY_PATH=${gdal_prefix}/lib:$LD_LIBRARY_PATH
export GDAL_DATA=${gdal_prefix}/share/gdal
gdalinfo --version

The last command should output something like the following:
GDAL 2.3.2, released 2018/09/21.

Clean up: Navigate to repository and remove downloaded files.

cd <path-to-repo(sparse_image_fusion)>
rm -r <chosen-path-to-thirdparty-library-dir>/downloads

Link libraries to repository's main directory

ln -s <chosen-path-to-thirdparty-library-dir> <path-to-repo(sparse_image_fusion)/lib>

Compile the program

Depending on the underlying maching this program is compiled and run on, you need to modify the Makefile by setting the compiler variable CXX to either mpic++ (PC) or mpiCC (server).

module unload mpi.ibm           # <- on SuperMUC only
module load mpi.intel/2018_gcc  # <- on SuperMUC only
module load gcc/8               # <- on SuperMUC only
make

Run the program

Run the script run.sh, in which all program arguments are set including specification of paths to the input data set.

bash run.sh

On the SuperMUC, corresponding LoadLeveler scripts need to be created and submitted.

Flowchart

A technical flowchart describing the J-SpareFI-HM algorithm is given below.

claas-grohnfeldt/sparse_image_fusion