Classifying areas in scanned maps (not by colors but by pattern)

I have old land-use maps in a scanned format. I am trying to obtain proper land-use maps from it, similar to what is decribed here: Comparing areas in scanned historical land use maps using QGIS?. The problem I'm facing is that some of the land-use classes are drawn as pattern (e.g. forests below are green with sprinkled black dots). For many techniques in GIMP, such as selecting areas by colors / selecting all green pixels, which are in close to black pixels, the issue is the dithering and the labels/raster of the map. The colors are fairly easily recognized with any software.

My fairly general question is for tips on how to do this in ArcGIS/GiMP? I.e. possibly with the intermediate step of creating an image file, where the patterns are replaced by plain areas in high-contrast colors.

So far I've tried colour-filters and various ways of "selecting by color" in gimp.

Example from the map:

Fetzer's suggestion will probably work (using the supervised classification function) it is not so different in principle to handling sparse cloud cover.

If you have already tried this one could consider a local smoothing window to give the forest areas a more distinctive net colour, again followed by supervised classification. If you are lucky it might also turn the text into a distinct class that you can then combine with its respective background class. However depending on the precision needed for the outcome, it might be necessary to classify the other land covers separately and overlay them to retain edge precision elsewhere.

Specifically pattern based analytics e.g. ANN, are less well supported by ArcGIS, better to look at image segmentation software.

Watch the video: Section 4 (October 2021).