ECW, GetGeoTransform returns posetive fCellIncrementY

[llo@…]

Trying to display some sample ECW datasets, using Mapserver/GDAL failed, but 
the ECW files worked with some mapInfo product using a .tab file with 
georefrence info.

The problem was that the images was ignored due to a posetive fCellIncrementY 
returned from GetGeoTransform.

The following fix, was done on the GDAL CVS version from 16 June 03.

I added the following code to ECWDataset::GetGeoTransform, line 466
    if (padfTransform[5] > 0.0)
      padfTransform[5] = -padfTransform[5];
And
    if (adfGeoTransform[5] > 0.0)
      adfGeoTransform[5] = -adfGeoTransform[5];
To GDALDataset *ECWCreateCopy, before line 671.

The the problem was solved. I dont know if this is a problem in the ECW driver, 
or if the problem is caused by a malformed ECW georeferencing info.

[warmerdam]

Lars,

I have reviewed a bunch of georeferenced ecw files I have, and they all have
negative fCellIncrementY values.  I am therefore hesitant to apply your change
for fear of messing up the expected behaviour in cases of ungeoreferenced 
images.  

So, for now I will do nothing, but if I see it coming up alot I will apply
the suggested patch. 

Thanks,

[gaopeng]

I reopen this ticket as we have many cases where fCellIncrementY has positive value. We have cases reported by the following offices in different regions:

GISTEC, SHARJAH ESRI KOREA INC, SEOUL ESRI CANADA LIMITED, TORONTO ESRI AUSTRALIA PTY LTD, BRISBANE LOCUS LTD, DEVONSHIRE (Bermuda)

I'd suggest the following fix, which should be safer:

        adfGeoTransform[5] = -fabs(psFileInfo->fCellIncrementY);

[warmerdam]

Gao,

Can you provide a sample file? Do you have any idea why some files have a positive fCellIncrementY? I'd like to take the issue up with ERDAS to see if they have any advice.

[gaopeng]

I uploaded one sample at  ftp://GDAL@ftp.esri.com/ECW/. I can get more if needed.

[warmerdam]

I have examined the file, and unfortunately it is not practical to do any ground truthing to determine if flipping the sign of the y cell increment would give a proper result. I have emailed Erdas to see if they have any advice on the topic.

[warmerdam]

Here is my understanding of how Erdas Imagine use this parameter. If the ecw file is raw (no map-projection info), then fCellIncrementY is positive (i.e., pixelSize->height) If the ecw file has map-projection info, then fCellIncrementY is negative (i.e., -pixelSize->height)

Best regards! Haiyan Qu

[warmerdam]

Based on this, I'm going to apply the proposed patch. I can confirm that the file in question is RAW; however, it has an origin so it seems likely that it is in some coordinate system but the coordinate system is just not described in the file. I'm concerned that flipping the sign might still not place the file appropriate.

There is also a risk that this will screw up legitimately south up images though I can't say that I've knowingly encountered such a thing.

[warmerdam]

As I'm tentative about this fix, I have only placed it in trunk (r21598) so far. Do you need this in 1.6-esri?

[gaopeng]

Yes. Please.

[warmerdam]

I have pushed the change into 1.6-esri (r21592) and 1.8 (r21593).

I'd appreciate hearing if it causes problems for other files.