[gdal-dev] GUI for gdalwarp ?

[Balint Cristian]

> On Thu, Jun 26, 2008 at 09:15:56AM -0400, Frank Warmerdam wrote:
>> Jean-Claude Repetto wrote:
>> >Hi,
>> >
>> >Is there a GUI that allows entering GCPs graphically and then calls
>> >gdalwarp to calibrate a map ?
>> 

I am working on a robust CLI tool based over
gdal/geos library wich can do automated georeference of
cartographic raster maps which contain mercator
or any other border straight lines.
  Tool works simply, it do full-radon transform 
(a higly accelerated and optimized one) to extract 
straight lines, than got intersection points. Intersection 
points are corelated  using 'ransac' scale/rotation invariant 
method against a generated grid which is generated out 
from knowing of raster map's name/position, and resulted 
valid points are inserted into raster as gcp.
  Advantage is that it can do mercator grid reference 
instead of just 4 corner points only one, and is fully
automated.

 Preliminary results showing how accurate is radon 
transform, and its complete noise-immunity:

http://openrisc.rdsor.ro/radon-020k.TIF

   However code is not complete yet, but i assume next month
will be, including extensive usage documentation and description how
it works, with math fundaments too. I got impressive results so far, 
basicaly on a sample of random ~1000 raster maps not a single 
line was missed  or error induced, radon simply do an amasing
job to detect any straight line, no mather how much noise we have
or how poor are the lines due to bad low quality scans.

  Project is intended as a free one, its subject of my PhD thesis,
i will happily publish it as i will done it, code its simple and not exceed
2k lines, but the math behind is the real killer part.
 I hope it will get some interest from peoples around the world
but lets see, i have no time set up a wiki to document development 
process and my intention i better focus to finish it rather than 
form a small team or wiki pages whatever.


 More details on radon:
http://eivind.imm.dtu.dk/staff/ptoft/Radon/Radon.html

 However my merit in this work is the accelerated radon, 
its a difference to do 30hour over 20k*20k pixel or just 15
minutes, plus corelated points should perfect valid/invalid match
using a modified ransac algorithm.

//cristian