[GRASS-dev] Introduction and GSoC

[Moritz Lennert]

On 03/04/12 10:23, Moritz Lennert wrote:
> On 02/04/12 22:03, Markus Neteler wrote:
>> On Mon, Apr 2, 2012 at 1:42 PM, Moritz Lennert
>> <mlennert at club.worldonline.be> wrote:
>> ...
>>> 4. Currently GRASS does not provide any image segmentation as such.
>>> i.smap
>>> contains image segmentation in its process, but the user cannot get
>>> segmented outputs.
>>
>> Just a small comment - even not this one?
>> http://grass.osgeo.org/wiki/GRASS_AddOns#r.seg
>> r.seg performs image segmentation and discontinuity detection (based on
>> the Mumford-Shah variational model).
>
> Ah, another GRASS module I didn't know about. ;-)
>
>>
>> Perhaps it could be extended?
>
> I'll have a look at it.

I did a rapid test with a landsat image:

for lambda in 0.01 0.1 1 10 100
   do
     for alpha in 0.01 0.1 1 10 100
       do
          r.seg in_g=L72199024_02420020729_B80 at PERMANENT 
out_u=test_seg_$lambda\_$alpha out_z=test_seg_dis_$lambda\_$alpha 
lambda=$lambda alpha=$alpha --o
          r.colors -e map=test_seg_$lambda\_$alpha color=grey
          r.colors -e map=test_seg_dis_$lambda\_$alpha color=grey
       done
   done

The lamba=100 and alpha=100 seems to give visually me roughly what I was 
thinking of, i.e. a series of regions. However, each pixel still has a 
different value and so I would have to go through more steps before I 
could try to delineate actual polygons which I could then classify.

AFAIU (and this is still a bit limited) r.seg seems to be more oriented 
towards visualisation than towards further treatment of identified 
"segments".

Moritz