Hyperspectral Fourier-Mellin algorithm (HYFM) on GPU

Requirements

• NVIDIA GPU with compute capability 6.1.
• Ubuntu 14.04 or later

Dependencies

• NVIDIA CUDA Toolkit
• Install: sudo apt-get install libmatio-dev

Usage

Run ./HYFMGPU with the following parameters:

./HYFMGPU <reference_image.mat> <target_image.mat>

Outputs:

• Textual result in console: scale, angle of rotation, and shifts over x and y axis.
• An original band of reference and target images in PGM format.
• A registered band image in PGM format.
• Registered hyperspectral image in MAT format.

Additional program

rot_scal allows generating new target images with specific a scale and angle of rotation for testing the performance of the method.

Use:

./rot_scal <image.mat> <scale> <angle>

The output image is saved in the same directory with the name rot_scal_out.mat.

Examples

Using Santa Barbara Line images:

./HYFMGPU SantaBarbaraLine2013.mat SantaBarbaraLine2014.mat

Using Pavia University image:

1. ./rot_scal PaviaUniversity.mat 2.0 90
2. ./HYFMGPU PaviaUniversity.mat ./rot_scal_out.mat

Citation

This algorithm was presented in the paper "GPU Accelerated FFT-based Registration of Hyperspectral Scenes" in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. If you use this program in your research projects, we encourage you to please cite our work:

@article{alvaro2017hyfmgpu,
    title = {{GPU} Accelerated {FFT-based} Registration of Hyperspectral Scenes},
    journal = {IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
    year = {2017}, 
    volume = {10}, 
    number = {11}, 
    pages = {4869-4878}, 
    author = {Ord{\'o}{\~n}ez, {\'A}lvaro and Arg{\"u}ello, Francisco and Heras, Dora B},
    doi={10.1109/JSTARS.2017.2734052}
}