Software for Hyperspectral Super-Resolution with Coupled Tucker Approximation

Copyright (c) 2019 Clémence Prévost, Konstantin Usevich, Pierre Comon, David Brie
Contact: clemence.prevost@univ-lorraine.fr

This MATLAB software reproduces the results from the following article:

@article{prevost2020hyperspectral,
  title={Hyperspectral super-resolution with coupled Tucker approximation: Recoverability and SVD-based algorithms},
  author={Pr{\'e}vost, Cl{\'e}mence and Usevich, Konstantin and Comon, Pierre and Brie, David},
  journal={IEEE Transactions on Signal Processing},
  volume={68},
  pages={931--946},
  year={2020},
  publisher={IEEE}
}

Link to the project

Content

• demo.m: demo file with minimal requirements

• /demos : contains demo files that produce tables and figures (including main.m

• /data : contains data for synthetic examples (Section VI.D)

• /figures : where the tables and figures are saved

• /images : contains illustrative figures for this README.md

• /src : contains helpful files to run the demos

Minimal requirements

In order to run the demo file demo.m, you will need to:

• Download and install Tensorlab 3.0: https://www.tensorlab.net
• Download some hyperspectral data from http://www.ehu.eus/ccwintco/index.php/Hyperspectral_Remote_Sensing_Scenes and load them into MATLAB

Full requirements

In addition, in order to run the demo file demos/main.m, you will need to:

• Download codes by C. Kanatsoulis from https://github.com/marhar19/HSR_via_tensor_decomposition and add them to your MATLAB path
• Download codes for HySure from https://github.com/alfaiate/HySure and add them to your MATLAB path
• Download codes for FUSE from https://github.com/qw245/BlindFuse and add them to your MATLAB path

Demo file

A demo with minimal requirements is available. To proceed, please run the demo.m file.

Load hyperspectral data

Start by loading hyperspectral data. This parts manually generates the HSI and MSI from the groundtruth SRI, by computing the degradation matrices. Below is an example of a single spectral band for the HSI and MSI from the Indian Pines dataset.

Run algorithms

The next step is to run the algorithms SCOTT and BSCOTT (Blind-SCOTT).

R = [40 40 6]; 
[SRI_hat1, ~] = scott2(HSI, MSI, P1, P2, Pm, R);
opts.Nblocks = [4 4];
[SRI_hat2, ~] = bscott(MSI,HSI,Pm,R,opts);

The algorithms return the estimated SRI of given multilinear rank R. For BSCOTT, the HSI and MSI are split into corresponding opts.Nblocks blocks along the spatial dimensions. Note that BSCOTT does not require the knowledge of P1 and P2.

Comparison metrics

You can then generate a table of comparison metrics and computation time:

Algorithm	R-SNR	CC	SAM	ERGAS	Time (s)
SCOTT	26.3908	0.887454	2.32401	1.0587	0.26491
BSCOTT	18.647	0.820201	4.27434	2.62442	0.1644

Reproduce figures and tables from the paper

To do so, you need to run the demos/main.m file. Here, a menu is available and allows you to choose which figure or table you want to generate. Each number in the table below corresponds to a set of figures, and/or tables.

Number	Content
1	produces Tables I and II and Figures 3 and 4
2	produces Tables III and IV and Figures 5 and 6
3	produces Table V
4	produces Figures 7 and 8
5	produces Figures 9 and 10
6	produces Figure 11
7	produces Figure 12
8	produces Figures 13 and 14
9	produces Figures 15 and 16
10	produces Figures 17 and 18
11	produces Figures 20, 21 and 22