Scanning maps into GRASS

[charnotw]

On Tue, 23 Jul 1996, R.A.Sanderson wrote:

> Does anyone have experience of using a flatbed scanner (e.g. HP ScanJet) to
> scan colour maps from paper, and convert them into GRASS raster maps
> (running GRASS on SunOS).  I have managed to get a simple test map,
> containing only 2 colours, into GRASS, via r.in.sunrast, but it has proved
> very difficult to clean up the image, to remove speckles without changing
> the total areas of each scanned colour.  I've tried using r.buffer and
> r.neighbours, together with r.mapcalc.  The main problems are associated
> with text on the map resulting in a "noisy" image, and editing out these
> points using d.rast.edit is too slow to be practical.
> 
I had a go using the image classifiers to clean up a soil map, each soil unit
was depicted on the map by a different colour, so it was just a case of
defining training areas.  Problems I encountered were to do with the dithering
used to obtain the colours on the map, folds in the map caused interference 
patterns when it was scanned, and the writing and other black lines tended 
to get in the way. I used r.mapcalc to split the map in to red green and blue 
bands, i.smap proved to be the best classifier and I found that if I scanned
the map twice with the scanner bulb at different brightnesses (so I had six
bands: red: light and dark, blue: light and dark and green: light and dark) 
I got the best results.  Even so the results were fairly mediocre.

BTW I found using i.rectify good for getting the maps into the right position, 
better than just editing the header as it is nearly impossible to get the maps
accurately lined up north-south-east-west in the scanner, they always tend to 
slip a bit.

cheers   Tom


Tom Charnock                       O--O
Dept Civil Engineering            (~~~~)
Aston University                 (  __  )
Birmingham B4 7ET UK             /|\  /|\
charnotw at sun.aston.ac.uk