[GRASSLIST:4892] conditional growing of a region

[Thomas Dewez]

Dear all,

I have been trying to regionalise a DEM into particular landforms. In 
this instance, I am after identifying terraces. My original attempt was 
with a classification type method whereby I identified threshold in 
variables such as slope gradients and flatness. While this is one 
way of dealing with this question, I am a little disappointed by the 
results because they don't respect the visual geomorphological 
interpretation of the landscape. It is striking that terraces are chunks 
of land with a particular aspect and slope value (shallow slope 
obviously), yet thresholding the slope and aspect ignore the 
topological adjacency of pixels. Therefore, I'd like to build an 
algorithm that would work like the "magic wand" in photoshop. That 
is a method that looks at the characteristics of a seed point and test 
the membership of adjacent pixels according to some decision rule.

In practical terms, how does one go about comparing each pixel of 
a buffer zone to a seed point. I found the r.grow module which will 
do to expand the seed point by one pixel in all direction, but once 
this ring of neighbours is determined, how does one test them one 
by one? Or is there more nifty trick to deal with this?

Thanks for your help,

Thomas

Thomas Dewez
PhD Student
Dept Geography & Earth Sciences
Brunel University (West London)
Uxbridge UB8 3PH, Mddx
United Kingdom

Phone: +44-(0)1895-203215
Fax:     +44-(0)1895-203217

e-mail: thomas.dewez at brunel.ac.uk