The Coregistration Swiss Army Knife: GeFolki

GeFolki is a coregistration software that has been developped in the framework of the MEDUSA project, first for SAR/SAR co-registration, then for other cases of remote sensing image coregistration, including hetergeneous image co-registration (ex LIDAR/SAR, optics/SAR, hyperspectral/optics and so on)

Detailed instructions for use.
Notice en français.

 

 


We hope this code will be helpful and we remain to your disposal and we are interested in getting back your remarks.

If you are using the result of GeFolki in your project, we kindly ask you to cite

– for SAR/SAR coregistration (interferometry, change detection, and so on) [Reference1]

Plyer, A., et al. (2015). A New Coregistration Algorithm for Recent Applications on Urban SAR Images. Geoscience and Remote Sensing Letters, IEEE, 12(11), 2198-2202.

– for other cases (optics/SAR, optics/hyperspectral, LIDAR/SAR, etc.) [Reference2]

Brigot, G., et al. (2016)  Adaptation and Evaluation of an optical flow method applied to co-registration of forest remote sensing images, accepted with modifications in  IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,  Volume: 9, Issue7, July 2016

We are also interested by your feedback or questions about GeFolki.

Please contact

 


You can DOWNLOAD the source code in a matlab .zip directory under a GPL license.

After decompression, this directory contains:

– a directory “datasets” that are used in the demonstration file “main.m”. Origins of the different datasets are described in the readme.txt

– the different matlab source codes. The demonstration are proposed in the main.m file.

For information, the execution time of the whole script main.m with 3 examples of co-registration is 90 seconds on a Intel® Xeon(R) CPU E5-1603 0 @ 2.80GHz × 4

You can also download the different files in https://github.com/aplyer/gefolki

 


All data provided in this Datasets to test GeFolki come from various institutions:

The SAR and DEM data acquisition has been funded by ESA through the BIOSAR 3 campaign.

– Polarimetric airborne SAR data was acquired by ONERA under ESA Contract 4000102285 “Technical Assistance for the Development of Airborne SAR and Geophysical Measurements during the BioSAR 2010 Experiment”. The final report is available [here]

– The initial LIDAR image used in this set is a 0.5×0.5m resolution DEM processed by SLU. Thanks to Johan Fransson from SLU (Swedish University of Agricultural Sciences),

and by Lars Ulander from FOI (Swedish Defence Research Agency)

– The optical image over Sweden has been provided by Lantmäteriet. It has been processed from a RGB orthophoto initial dataset (GSD-orthophoto 0.5m RGB product)

 


More Explanations About GeFolki

The coregistration of images is based on three steps:

– Initialization: we choose a common resolution grid to surimpose the images. This can be done by a common geocoding process, but in the case of geocoding, the accuracy of coregistration will be highly dependent on the availability of a DEM and on its precision, as well as the orbit parameters, and this step is not enough.

– Computation of the residual existing distortion between two images. This is the Heart of GeFolki.

– Re-sampling

Non-rigid image registration is widely investigated in various fields beyond the scope of remote sensing, such as computer vision and medical imaging. The methods differ to fit these different constraints: robustness and speed of execution, or differing nature of the images to register.

Most non-rigid registration methods are based on a similarity function that is optimized to find the underlying deformation. Mutual information is the most widespread similarity measure for registration of multimodality images. However, mutual information is quite time consuming, and therefore, is generally preferred for sparse motion estimates, because would be very long if applied to all pixels of the images.

In the field of computer vision, there are also coregistration methods known as optical flow methods, that are non-parametric and dense methods. They have often been developed in a video context where intensive efforts have been led in producing algorithms that combine robustness, precision and high computing speed. However, most optical flow methods rely on an observation model where intensity remains the same from one image to another. This hypothesis is called the brightness constancy.

GeFolki (Geoscience Extended FlOw Lucas- Kanade Iterative) is an adaptation of an optical flow method [1], that can handle multiple remote sensing images.

It is already a bold work to apply such an optical flow algorithm to radar images, knowing that they contain speckle, which may undermine the adequacy of the brightness constancy constraint. However, since adequate settings managed to make it effective all the same, we also have developed new processing steps to adapt our method to a more tricky case: the heterogeneous images. The brightness constancy model does not apply at all in this case. Therefore, making this kind of optical flow method compatible anyway with the obvious changes in existing brightness proves to be a major challenge.

To overcome this difficulty, algorithmic modules have been developed to be applied to heterogenous image intensities, to ensure coregistration while maintaining robustness and computational efficiency. Several success cases have already been obtained:

–    About SAR coregistration [2]: for interferometric or non interferometric conditions, between different SAR modalities and different sensors, at different wavelengths, both for forest or urban media.

–    About heterogeneous image coregistration [3]: for LIDAR/SAR, optical/SAR, hyperspectral/SAR, hyperspectral/LIDAR

Current limitations in dealing with each specific situation are currently investigated.

[1] Le Besnerais, G. and Champagnat, F. (2005)  Dense optical flow by iterative local window registration, in Image Processing, ICIP 2005. IEEE International Conference on, vol. I, pp. 137–140, Sept. 2005.

[2] Plyer, A., et al. (2015). A New Coregistration Algorithm for Recent Applications on Urban SAR Images. Geoscience and Remote Sensing Letters, IEEE, 12(11), 2198-2202.

[3] Brigot, G., et al. (2016)  Adaptation and Evaluation of an optical flow method applied to co-registration of forest remote sensing images, accepted with modifications in  IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,  Volume: 9, Issue7, July 2016

Attached Files
# File Type File Size Download
1 .pdf 1.07 MB anewcoregistration
2 .pdf 8.59 MB gefolki
3 .zip 94.78 MB GeFolki_v1.0
4 .pdf 65.24 MB BIOSAR_2010_FinalReport_v1.0