[GRASS-dev] GSoC proposal "Implement cutline generation" - Looking for a mentor

Taïs Grippa tgrippa at ulb.ac.be
Fri Feb 3 01:31:06 PST 2017


Dear dev team,

I would like to propose an idea for the 2017 GSoC (it is already on the 
GSoC 2017 GRASS wiki page).

It will be focused on 'meaningful' cutline generation.  I recently read 
a paper describing a method for cutline generation (see below [1] for 
the article, [2] for more references).

The procedure is composed of the following main steps:
1) Apply a High-pass filter - the "Prewitt South" filter 
(https://en.wikipedia.org/wiki/Prewitt_operator) to produce an edge image;
2) Apply a low-pass filter (mean for example) on the previous result;
3) Find the starting and finishing points (just based on pixel values in 
the first/last column/row);
4) Create an adjacency matrix on the result of 2) and use weights to 
give higher costs for specific directions (and force the cutline in to 
be quite linear between starting and finishing points);
5) Use the Dijkstra's algorithm 
(https://en.wikipedia.org/wiki/Dijkstra's_algorithm 
<https://en.wikipedia.org/wiki/Dijkstra%27s_algorithm>) to find the 
optimal cutline with a lowest cost between the starting and finishing point

Finding meaningful cutline in the image could be very interesting for 
tiling images in order to process segmentation in parallel. Instead of 
using arbitrary linear tiling, the segmented tiles could then be merged 
without post-processing of segments touching the tile's border.

If this algorithm would be implemented during the GSoC, it could be 
further adapted for huge enhancement of i.image.mosaic to enable 
mosaicking of overlapping scenes using cutline and become a real 
alternative to proprietary software 
(http://gis.stackexchange.com/questions/127310/how-to-create-a-mosaic-in-qgis-with-cutline-and-feathering-for-landsat-8-imagery). 


As the proposed solution [1] for cutline is working on a single image, 
It should be adapted to take into account pixel values of two images so.
I can give a support for the content of the project, but I don't have 
the necessary programming skills for supervision of GSoC student. So, I 
would like to know if someone would be interested by being the mentor 
for this project !
Concerning the language requirements for this project I guess it will be C ?

[1] Soares, Anderson Reis, Thales Sehn Körting, et Leila Maria Garcia 
Fonseca. 2016. « Improvements of the divide and segment method for 
parallel image segmentation ». Brazilian Journal of Cartography 68 (6). 
https://www.researchgate.net/publication/286929006_Improvements_of_the_divide_and_segment_method_for_parallel_image_segmentation

[2] Körting, Thales Sehn, Emiliano Ferreira Castejon, et Leila Maria 
Garcia Fonseca. 2013. « The Divide and Segment Method for Parallel Image 
Segmentation ». In Advanced Concepts for Intelligent Vision Systems, 504 
15. Springer, Cham. doi:10.1007/978-3-319-02895-8_45. 
https://www.researchgate.net/publication/265794792_The_Divide_and_Segment_Method_for_Parallel_Image_Segmentation

Best,
-- 

*M. GRIPPA Taïs*
Geographer - Phd student IGEAT - ANAGEO
ULB CP 130/03 Av. F.D. Roosevelt, 50 - 1050 Bruxelles
Tél: 02/650.68.06
Mail: Tais.Grippa at ulb.ac.be
Web: http://maupp.ulb.ac.be/
<http://maupp.ulb.ac.be/> 
<http://igeat.ulb.ac.be/fr/unites-de-recherche/details/unit/analyse-geospatiale/>

-------------- section suivante --------------
Une pièce jointe HTML a été nettoyée...
URL: <http://lists.osgeo.org/pipermail/grass-dev/attachments/20170203/d03da921/attachment-0001.html>
-------------- section suivante --------------
Une pièce jointe autre que texte a été nettoyée...
Nom: maupp_logo_only_header_small_2.png
Type: image/png
Taille: 4388 octets
Desc: non disponible
URL: <http://lists.osgeo.org/pipermail/grass-dev/attachments/20170203/d03da921/attachment-0001.png>
-------------- section suivante --------------
Une pièce jointe autre que texte a été nettoyée...
Nom: logo.gif
Type: image/gif
Taille: 871 octets
Desc: non disponible
URL: <http://lists.osgeo.org/pipermail/grass-dev/attachments/20170203/d03da921/attachment-0001.gif>


More information about the grass-dev mailing list