Variational fusion of hyperspectral and multispectral images

Docker

This project is coded with C++ and the libraries needed for its functionning are included in a Dockerfile which enables our code to work on any OS based systems

• First Docker needs to be installed on you system with the help of this link: https://docs.docker.com/engine/install/

• Clone our project in the location of your choosing on your local system and open a terminal in the cloned folder

Building the docker container

From the Dockerfile we're going to build an image called "hsmsfusion" in the folder downloaded from our Github repository

docker build -t hsmsfusion .

Fusion

At this point all the libraries and the packages are installed. The next commannd line does multiple manouvers: it starts a docker container named demofusion from the built image, it mounts a volume on the docker container that points to the demo folder in downloaded one and launches the demo.sh script.

The demo.sh script executes three commands: it moves to the demo folder, it creates the HS and MS images and all the data needed for the fusion and finally, it carries out the fusion with the generated data.

docker run --name demofusion -v $PWD/demo:/home/demo hsmsfusion sh demo.sh

Using other dataset

In case you want to run the fusion with your own hyperspectral (hs) and multispectral (ms) data, the interpolated hs image, the panchromatic image and the interpolated panchromatic image should be generated. For this, you just have to, in the batch_hyperspectral_data.sh, comment out the parts that generate the hs and ms images and rename the hs and ms images into: name.hyper.noisy.tif and name.multi.noisy.tif respectively where name is the name of the image before executing the demo.sh script.

Recovering the fusion result

The fusion result will be available in the demo folder in the cloned repository.

Fusion example

                       

Example of fusion on the image "Bicycles" from Harvard dataset. From left to right. Ground truth image, hyperspectral image and the fused image with the variational model.

Citation

For citing our work

@article{mifdal2021variational,
  title={Variational Fusion of Hyperspectral Data by Non-Local Filtering},
  author={Mifdal, Jamila and Coll, Bartomeu and Froment, Jacques and Duran, Joan},
  journal={Mathematics},
  volume={9},
  number={11},
  pages={1265},
  year={2021},
  publisher={Multidisciplinary Digital Publishing Institute}
}

Contact

If you have any questions please let us know: [email protected] and [email protected]