Re: Camera input profiling make/model handling

[Pascal de Bruijn <pmjdebruijn pcode nl>]

On Thu, Dec 3, 2009 at 8:09 PM, Richard Hughes <hughsient gmail com> wrote:
> 2009/12/3 Pascal de Bruijn <pmjdebruijn pcode nl>:
>> I haven't had time yet to properly check this, but what is used by GCM
>> for make/model info when profiling cameras/scanners?
>
> Information from sysfs.

Maybe I'll test this over the weekend...

>> At least for camera's it seems wise to use Make/Model from EXIF. This
>> way, software like UFRaw and DarkTable (and others) can use the EXIF
>> make/model to match which profiles should be valid for a certain RAW
>> file...
>
> GCM doesn't treat the device like a storage device, it treats it as a
> USB endpoint device, which means we don't look at the files contained
> within.
>
>> Luckily EXIF is much better human readable (and generally less
>> ambiguous) than EDID...
>
> I think this is the wrong layer. What happens if the EXIF data for one
> file on the "device" has different colorspaces or even two different
> device models? The device can carry all number of different EXIF
> files, but the device can only have one profile. If you inserted a
> pendrive of photos, would you expect to assign a per-pendrive ICC
> profile?

Huh?

A "Device" does not really have different color spaces... a camera's
sensor (RAW file) has a native color space characterized by the
profiling of the IT8 target (or another target, which is not really
relevant, it's a means to an end).

Any camera that produces sRGB or AdobeRGB, is likely to be not fit to
be really profiled. Since a camera most likely does not statically
convert sensor data to sRGB/AdobeRGB, it's likely a dynamic (image
dependant) process.

Anyway I filed a feature request with UFRaw, for "special" profiling
mode, to convert a camera RAW file an image with linear gamma in the
native camera color space:
  http://sourceforge.net/tracker/?func=detail&aid=2841222&group_id=127649&atid=709089

If one would have a pen drive of photos, the pen drive would be
completely and utterly irrelevant from a profiling/profile assignment
perspective. Each and every file on the drive could require a
different profile (at least for proper results). And normally EXIF
could be reasonably reliably used to "match" that.

> I think it's sane for the end application to open the file with EXIF
> metadata and then decided what to do. If it wants to just use the
> embedded profile, that's fine, or it can ask gcm for the default
> device profile from the volume (which it can get from udev).

Again, with images, only the image color space is truely relevant
(embedded or not). However, RAW files do not have assigned (embedded)
profiles. UFRaw/DCRaw "solves" this by ripping generic profiles (Color
Matrix) from Adobe DNG SDK, and applying them on a by EXIF make/model
basis.

Please do note, the above applies to camera's. I'm not really sure
about scanners.

Regards,
Pascal de Bruijn