[Qgis-user] Segmentation : how to add texture criterion ? (not only colour/spectral variable)

[Stephane Goldstein]

I had good results including texture as an input for Sentinel-2 image
classification.
It would probably bring you way more additional information on a very high
resolution image like SPOT

I used the GRASS texture tool. But there are so many outputs and parameters
that you can set, that you might quickly become overwhelmed with too many
data outputs.
To help with that, I ran a PCA transformation on the Sentinel-2 spectral
bands, then only ran texture on the PCA band that was most obviously
discerning the features I was interested in.

You might also have to reduce the radiometric resolution of the data on
which you want to calculate texture (e.g from the original 16 or 32 bits to
8 bits) otherwise your RAM is likely to explode when GRASS calculates the
grey level co-occurrence matrix.
Good luck


On Tue, 4 Mar 2025 at 22:06, celati Laurent via QGIS-User <
qgis-user at lists.osgeo.org> wrote:

> Hello.
> My goal is to generate a relevant segmentation that would delimit the main
> physiopnomic units (homogenous areas). Up to now, I have done several tests
> with qgis and orfeotoolbox 9 with the OTB Generic Region merging module : GenericRegionMerging
> — Orfeo ToolBox 7.0.0 documentation
> <https://www.orfeo-toolbox.org/CookBook-7.0/Applications/app_GenericRegionMerging.html>
> with those folowing settings
>
>    - input image : spot 6/7 satelite image (band blue, green, red, near
>    infrared)
>    or Aerial french mapping agency images (50 cms bd ortho with green,
>    red and near infrared).
>    *Homogenity criterion to use : Baatz & Schape
>    - Threshold for the criterion : 500
>    - Number of iterations`: 0
>    - Weight for the spectral homogeneity : 0.1
>    - Weight for the spatial homogeneity : 0.1
>
> The result is rather interesting for certain areas, in particular crops
> areas, without too much vegetation. on the other hand, the method does not
> seems optimal, in particular for forest areas.
> I’m taking the liberty to paste some screenshots :
> the first SS showing interesting result on crop areas :
>
> [image: 4d3f841e567bd5ac5ce37035342c4e71e78d16c6.jpeg]
> A second SS showing the non relevant result on forest areas :
> [image: 9b64943de826eddd230762b3d2a5eb358cc423ec.jpeg]
> Perhaps for these forest areas, the color/spectral criterion is not
> relevant. It would be better to generate / integrate texture information?
> As far as i understood, this GRM OTB module is based only on
> spectral/colour criterion but not take into consideration texture
> information.
> Are there other orfeotoolbox segmentation module that takes into account
> not only the color criterion (spectral) but also texture information?
> If this is not the case, should i create some kind of new raster (add a
> band to my existing color infrared raster) dedicated to texture and/or ndvi
> information?
> Thanks so much for your guidances.
>
>
>
> _______________________________________________
> QGIS-User mailing list
> QGIS-User at lists.osgeo.org
> List info: https://lists.osgeo.org/mailman/listinfo/qgis-user
> Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-user
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/qgis-user/attachments/20250305/992f0fe8/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 4d3f841e567bd5ac5ce37035342c4e71e78d16c6.jpeg
Type: image/jpeg
Size: 66056 bytes
Desc: not available
URL: <http://lists.osgeo.org/pipermail/qgis-user/attachments/20250305/992f0fe8/attachment-0002.jpeg>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 9b64943de826eddd230762b3d2a5eb358cc423ec.jpeg
Type: image/jpeg
Size: 109305 bytes
Desc: not available
URL: <http://lists.osgeo.org/pipermail/qgis-user/attachments/20250305/992f0fe8/attachment-0003.jpeg>