[GRASS-user] Classification of greyscale images

[Moritz Lennert]

Hi Paola,

Automatically classifying monochrome images is not easy. If you really 
want to use classical methods, your best bet is probably to create a 
series of pseudo-bands based on indicators such as texture (e.g. using 
r.texture, or possibly also r.neighbors). Another approach would be 
object-oriented, first segmenting the image into coherent segments and 
then calculating indicators for each of these segments, but segmentation 
generally also works much better on multiband imagery, so if you use, 
e.g., i.segment, then you would probably also have to create a series of 
pseudo-bands. Or you can try with r.clump.

When I've worked on such images in the past, neural network approaches 
seemed to actually work best. Here's an example on (fairly bad quality) 
historic images: 
https://dipot.ulb.ac.be/dspace/bitstream/2013/312410/3/FCN_for_historical_images_AUTHOR_VERSION.pdf. 
Since we've done that work, networks have evolved very rapidly, and you 
should probably be able to get better results nowadays.

Moritz

Le 4/07/24 à 18:23, Paola Salmona via grass-user a écrit :
> Hi everybody,
> I'm trying to classify some photograms from a Royal Air Force flight 
> in 1944. They are in greyscale and when I import them into grass using 
> r.in.gdal I get only one band and I can't use image processing command 
> such as i.cluster or i.gensig.
> I have tried to cheat by assigning to an image a color table, 
> exporting it, converting in RGB in an external software, then 
> importing it again into grass.
> It "worked" with some alerts, but, beside being a cumbersome method, I 
> am not sure about the quality of the re-imported image.
> Does anybody know a better way?
> Thank you very much!!!
> Paola
>
> _______________________________________________
> grass-user mailing list
> grass-user at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/grass-user
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/grass-user/attachments/20240705/c0405615/attachment.htm>