[Qgis-user] Georeferencer 3.1.9 bug (?) and work-around

[O]

Dear QGIS users,

Summary: I'm running the Georeferencer plugin 3.1.9 with QGIS 1.6.0 and it
doesn't work correctly -- the raster is distorted and plotted in the wrong
location. However, I found a work-around that fools the plugin into giving
correct results, and appears to point out a serious bug in the plugin. (Or,
I'm
doing something wrong -- please let me know.)

(And please tell me if I've posted this in the wrong place.)

Suppose I have an image called "input.tif" that has some geospatial metadata
with a CRS and image bounds. For the following discussion, here are the
relevant lines that result from running "gdalinfo input.tif":

Size is 1101, 1091
...
Upper Left ( 4578.469, -13361.035) ( 5d26'57.36"W, 2d10'18.28"S)
Lower Left ( 4578.469, -13652.219) ( 5d26'57.36"W, 2d10'35.98"S)
Upper Right ( 4872.322, -13361.035) ( 5d26'39.49"W, 2d10'18.28"S)
Lower Right ( 4872.322, -13652.219) ( 5d26'39.49"W, 2d10'35.98"S)
Center ( 4725.395, -13506.627) ( 5d26'48.43"W, 2d10'27.13"S)

Now suppose I open the Georeferencer and open this image, and define a
transformation (say, "polynomial 1") with three ground control points
(GCPs).
When I tell it to "start georeferencing" and to load the result into my QGIS
project, the image is hugely distorted and located in the wrong place. I
found
a BIG CLUE about what's happening when I asked it to "generate GDAL script".
The result was:

gdal_translate -of GTiff -gcp 4690.78 13510.5 4434.32 -13412.7 \
-gcp 4721.32 13469.4 4462.09 -13371.6 \
-gcp 4753.5 13539.4 4497.52 -13458.7 \
-gcp 4692 13536.2 4430.49 -13445.3 \
"/path/input.tif" "/tmp/input.tif"

gdalwarp -r cubic -order 1 -co COMPRESS=NONE "/tmp/input.tif" \
"/path/output.tif"

But if we look at the man pages for gdal_translate, the format for input of
GCPs
is:

-gcp pixel line easting northing elevation

The first argument to "-gcp" should be the x coordinate in *image* space,
not in
*coordinate* space. It should be the column number of the pixel in question
(I
don't know if it can be a floating point; hopefully it can). The second
argument to "-gcp" has to be the *row* number of the pixel in the image
being
georeferenced. The "easting" and "northing" are indeed the x and y
coordinates
in the target coordinate system. But notice that in the gdal_translate
command
that the Georeferencer generated, the first two arguments are positions in
coordinate space, not image space!

So, here's the work-around. We saw above that the size of the image is 1101
x
1091. All we have to do is change the bounds of the input image so that the
image's geographic coordinates match the image space coordinates. We can do
that easily with gdal_translate, being careful to account for QGIS using a
positive-up y-axis. Here's the command for the current case:

gdal_translate -of GTiff -a_ullr 0 0 1100 -1090 input.tif input_fixed.tif

Now, when we open the new input image (input_fixed.tif) into the
Georeferencer
and select GCPs, etc, it actually works!

M
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/qgis-user/attachments/20110608/97b50531/attachment.html>