My name is Tudor-Emil Coman and I study Computer Science at the Faculty of
Automatic Control and Computer Science in Bucharest, Romania as a MSc student.
I’m currently researching geospatial image processing technologies and I have
primarly worked with Sentinel 2 data on numerous frameworks like ESA SNAP,
GDAL, GeoDjango, PostGIS and of course GRASS GIS.
I’m very interested in working this summer for a project I have seen proposed
on the GSOC 2018 idea page regarding the development of image fusion algorithms.
An already existing fusion algorithm is the High-Pass Filter Addition technique
implemented in the i.fusion.hpf module [1]. The scope of the project would be
to expand the capabilities of GRASS to combine data from different sensors.
The first algorithm I would implement is based on Nicolas Brodu’s paper named
Super-resolving multiresolution images with band-independant geometry of
multispectral pixels [2]. The algorithm presented in this paper would increase
the resolution of certain low resolution bands by inferring some geometric
characteristics from high resolution bands. As opposed to the already existing
HPFA fusion method, this one does not require a panchromatic band making it
suitable for satellites that do not produce panchromatic bands.
I’m sure some other fusion algorithms are needed in GRASS-GIS that are not yet
implemented. I’m interested in suggestions about such algorithms so that I can
read about them and incorporate them into my final proposal.
On 02/03/18 10:35, Tudor-Emil COMAN (25633) wrote:
I'm very interested in working this summer for a project I have seen proposed
on the GSOC 2018 idea page regarding the development of image fusion algorithms.
An already existing fusion algorithm is the High-Pass Filter Addition technique
implemented in the i.fusion.hpf module [1]. The scope of the project would be
to expand the capabilities of GRASS to combine data from different sensors.
As note to other developers: Tudor has already sent me off-list a patch to i.in.spotvgt to respond to bug #3186 [1]. I still have to test it before applying, but if anyone else is interested, contact me.
The first algorithm I would implement is based on Nicolas Brodu's paper named
Super-resolving multiresolution images with band-independant geometry of
multispectral pixels [2]. The algorithm presented in this paper would increase
the resolution of certain low resolution bands by inferring some geometric
characteristics from high resolution bands. As opposed to the already existing
HPFA fusion method, this one does not require a panchromatic band making it
suitable for satellites that do not produce panchromatic bands.
I'm sure some other fusion algorithms are needed in GRASS-GIS that are not yet
implemented. I'm interested in suggestions about such algorithms so that I can
read about them and incorporate them into my final proposal.
Be sure to also look at the R-FUSE algorithm [2] mentioned on the GSoC ideas page.
In general, I think that general algorithms such as those two, that do not linked to sensor specificities would be great.
This said, IIRC i.pansharpen currently has limits in terms of applying the classical algorithms to images encoded in more than 8 bits. Improving this (implying changes to the underlying modules) might be another element of the GSoC project.
Please share your proposal as public, because you have submitted it as final, but we cannot either see it until deadline either comment it, so you cannot benefit from feedback and we will evaluate it as it is. Also, please read the recommendations at https://wiki.osgeo.org/wiki/Google_Summer_of_Code_Recommendations_for_Students and introduce yourself in soc mailing list.