Hello,
Some of my colleagues use eCognition for remote sensing data classification. After having done so for teaching, we are now looking into the possibilities of replacing proprietary with open source solutions in research.
eCognition uses an object-based approach to classification as opposed to pixel-based. As they put it in their brochure:
"The technology examines pixels not in isolation, but in context. It
builds up a picture iteratively, recognizing groups of pixels as objects.
Just like the human mind, it uses the color, shape, texture and size of
objects, as well as their context and relationships to draw the same
conclusions and inferences that an experienced analyst would draw."
Is something like this possible in GRASS ?
Or are there any other open source solutions using this approach ?
Moritz
The i.smap module (
http://grass.itc.it/grass63/manuals/html63_user/i.smap.html ) can be
utilised for image segmentation. So I suppose you can use the segments for
an object-based classification (!).
More details in the manual or from the experts... !
Nikos.
Moritz Lennert-2 wrote:
Hello,
Some of my colleagues use eCognition for remote sensing data
classification. After having done so for teaching, we are now looking
into the possibilities of replacing proprietary with open source
solutions in research.
eCognition uses an object-based approach to classification as opposed to
pixel-based. As they put it in their brochure:
"The technology examines pixels not in isolation, but in context. It
builds up a picture iteratively, recognizing groups of pixels as objects.
Just like the human mind, it uses the color, shape, texture and size of
objects, as well as their context and relationships to draw the same
conclusions and inferences that an experienced analyst would draw."
Is something like this possible in GRASS ?
Or are there any other open source solutions using this approach ?
Moritz
_______________________________________________
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user
--
View this message in context: http://www.nabble.com/object-based-classification-(vs-pixel-based)-tf4856980.html#a13898688
Sent from the Grass - Users mailing list archive at Nabble.com.
On 22/11/07 16:46, Nikos Alexandris wrote:
The i.smap module (
http://grass.itc.it/grass63/manuals/html63_user/i.smap.html ) can be
utilised for image segmentation. So I suppose you can use the segments for
an object-based classification (!).
i.smap might be a step in the right direction. I think (but am really not an expert) one of the important elements of the eCognition approach, is this:
Just like the human mind, it uses the color, shape, texture and size of
objects, as well as their context and relationships to draw the same
conclusions and inferences that an experienced analyst would draw."
So (just thinking out loud), maybe one could create a training map with a farily high number of examples of the types of objects one is looking for, then submit this to i.smap, clump neighboring cells together with r.clump. This would then be something like the "objects". Then you could calculate a series of indicators for each of these objects (shape, texture, etc) using r.texture, r.le.*/r.li.* (?) etc, on the original RS bands and then submit all the resultant raster maps based on these objects to a classification...
However, IIUC, eCognition actually builds the objects already using these object criterions, probably using some algorithm which goes through different possible objects and then tests them against the object criteria to decide which objects to create.
Carlos, I'll look into terralib and see what is in there. Thanks for the hint.
Moritz
Moritz
Markus once wrote in a related thread [1]:
"one could think of combining the (internal) results of
i.smap with the new DB engine to create a hierarchial model
with sets of rules. The pieces are partially there, but it
needs certainly some efforts. But I am not too familiar with
eCognitions approach."
Hope this is some information.
Maciek
[1]http://intevation.de/rt/webrt?serial_num=3898
do you have realize the objected classification ？if so ，can you share it with
us？
--
View this message in context: http://osgeo-org.1803224.n2.nabble.com/object-based-classification-vs-pixel-based-tp1880389p6354341.html
Sent from the Grass - Users mailing list archive at Nabble.com.