[PATCH] JPEGCreateCopy and 1 band palette indexed

[Even Rouault]

JPEGCreateCopy only checks for the number of source bands being 1 or 3 and maps 1 band to JCS_GRAYSCALE and 3 bands to JCS_RGB. This is really a minimal check. In the case we have only 1 band but whose color interpretation is GCI_PaletteIndex, we probably prefer to expand it as JCS_RGB rather than JCS_GRAYSCALE.

The enclosed patch implements this.

[warmerdam]

Even,

It is not normal practice for drivers to do paletted to RGB conversion on the fly. I believe a more correct (though less useful) behavior is for the driver to issue either a warning or an error for paletted images. A warning should be issued if bStrict is FALSE, and an error if bStrict is TRUE. If an error is issued, the translation should terminate.

If you feel strongly about drivers doing on-the-fly palette expansion, then I would suggest you raise it as a discussion topic on the mailing list.

[Even Rouault]

Yes, I was expecting that this patch is arguable.

I shoud have asked first if there is a way with gdal utilities to do paletted to RGB conversion ? I'm not aware of one.

Another option, that could be probably generalized to other drivers (I can't think of other examples though) , would be to add a creation option like "CONVERT_PALETTE_TO_RGB_IF_NECESSARY".

[warmerdam]

Even,

The pct2rgb.py utility can be used to convert pseudocolored images to rgb 24bit 3 band images.

Rather than having drivers do the expansion, I'd rather add options to gdal_translate to do it outside the driver.

[Even Rouault]

This is related to #1692

[Even Rouault]

I've commited in r12377 a simple patch that just warns the user that the source raster band is going to be considered as grey level.

#1692 suggests that this should probably be done in other drivers, but I can't test ECW or JP2K, so I let this one open