While We're on the Subject of DNGs . . .

[digitaldog]

http://www.color.org/ICC_white_paper_20_Digital_photography_color_management_basics.pdf
See anything there that refers to bit depth or encoding?

[Doug Gray]

You have not quoted what I wrote about ACR. Are you disputing Eric Chan's explanation (an Adobe engineer) provided in the link I posted about how ACR treats integer vs. floating point encoded files?

Again, I misspoke when I said it's 32 bit vs 8/16 bit. I misremembered, it's integer vs float.

Still think there is confusion related to gamma. Floating point images are in gamma=1. Integer coded images are usually in a higher gamma to prevent banding at low luminance. Especially 8 bit images. However, they can be converted to gamma=1 integer images.

Both Photoshop and Lightroom can use 32 bit floating images which are always in gamma=1. RAW files, before developing and outputting are essentially scene referred and become output referred when printed. One problem with Lightroom is that it simply isn't possible to print a scene referred image from Lightroom. It requires either Relative Colorimetric or, more usually, Absolute Colorimetric and also requires there be no BPC. Lightroom does not offer that.

A standard, output referred, Adobe RGB image, converted to 32 bit floating point and saved as a tiff, will be converted to gamma=1 and read in Lightroom using the same path as RAW files. However, it remains output referred. An output referred image cannot be converted to scene referred.

[Aram Hăvărneanu]

I think you’re confused between what scene referred is versus encoding.

I'm not. I think you are confusing what data means (something the author of the file intends) with how software interprets it in some case versus some other case (which might differ from the intention of the file author, if care is not taken).

You can have scene-referred data encoded as 8 bit integer, 16 bit integer, 32 bit integer, 16 bit float, 32 bit float, 64 bit float or any other format. Just as you can have an output referred image encoded exactly the same way, 8 bit integer, 16 bit integer, 32 bit integer, 16 bit float, 32 bit float, 64 bit float or any other format still.

In raw files, ACR treats all integer data as scene-referred. In TIFF files, it does not treat integer encoded data as scene referred by default (though you can make it so). In floating point encoded TIFF files though, by default, it treats the data as scene referred (even if, for some reason, it isn't actually scene referred e.g. Doug's example above).

[Aram Hăvărneanu]

Doug, we are in agreement completely. There is no confusion.

A standard, output referred, Adobe RGB image, converted to 32 bit floating point and saved as a tiff, will be converted to gamma=1 and read in Lightroom using the same path as RAW files.

Yes.

However, it remains output referred.

Absolutely correct.

An output referred image cannot be converted to scene referred.

Completely true.

We are in agreement here, not disagreement!

[digitaldog]

You are confusing gamma=1 colorspaces with scene referred. The two are unrelated.

Indeed. Bit depth too. And it appears that we are now all in agreement?

[digitaldog]

Once an image is output referred, you can't, as a practical matter, construct a scene referred image from it.

And when you use ACR as a filter in Photoshop, the output referred image you load is as you point out, is still output referred. ACR may 'treat them" in a linear processing color space, but they are output referred going in and out.

[digitaldog]

And when you use ACR as a filter in Photoshop, the output referred image you load is as you point out, is still output referred. ACR may 'treat them" in a linear processing color space, but they are output referred going in and out.

Plus, you don't even have to use ACR as a filter, one can easily open an output referred JPEG in ACR:

[BJL]

This discussion raised a question that I then asked in the beginner's form at
https://forum.luminous-landscape.com/index.php?topic=127201.msg1073969#msg1073969
but maybe that was the wrong place; I am sure that some people in this thread could answer it, so please help!

[Bart_van_der_Wolf]

This discussion raised a question that I then asked in the beginner's form at
https://forum.luminous-landscape.com/index.php?topic=127201.msg1073969#msg1073969
but maybe that was the wrong place; I am sure that some people in this thread could answer it, so please help!

Hi,

Demosaicing of an output pixel involves multiple surrounding photosites, also with different CFA filter colors.

The demosaicing process can get even more complicated when things like CA removal at the Raw level is included, or (geometrical lens) distortion. corrections, or diffraction correction.

Cheers,
Bart

[Jack Hogan]

in converting from Bayer CFA data to JPEG (or TIFF), is the value for a color channel at a pixel for which that photosite measures based just on the value for that photosite, or is there still interpolation, or are there various options in use?

For example: for a photosite that has a red filter, is the output file's red value at the corresponding pixel based only on that one measurement, or on some average also using data from other nearby "red photosites"?

As Bart says all nearby pixels are more or less correlated - no matter the color filter they sit under - and good demosaicers exploit this fact to their benefit.  You can actually 'see' the correlation in neutral portions of raw captures which, once properly white balanced, in effect produce a full resolution image without 'demosaicing'.  In such situations there is no resolution penalty for trichromacy, something that is by the way also true of the human visual system.

Jack