Ad
Ad
Ad
Pages: « 1 [2]   Bottom of Page
Print
Author Topic: Combining RAW + JPEG  (Read 13538 times)
Peter_DL
Sr. Member
****
Offline Offline

Posts: 421


« Reply #20 on: March 27, 2008, 02:52:24 PM »
ReplyReply

Many thanks for explanations and for your comments.
Largely agreed and/or accepted so that it would be redundant to quote everything.

Quote
... However, there's certainly an argument in favor of getting better starting points to help users improve their workflow. .
[{POST_SNAPBACK}][/a]
Yep.


Quote
I don't know about 'easier'. I guess it would depend on how easy that 'look' information is to access. The problem if you just treat DPP as a black box and grab the final image as a starting point is that, well, you're no longer starting from a raw file! You're editing beginning with an output-referred image in which case some of the raw benefits would be lost.
[a href=\"index.php?act=findpost&pid=184611\"][{POST_SNAPBACK}][/a]
Without intending to insist, the initial idea was as follows:

Referring to Camera Raw and the processing stage of linear-gamma ProPhotoRGB (1.0 pRGB) which ACR was reported to use as an intermediate working space, there should be access to scene-referred image data, freshly derived from “native” Raw by basic operations such as demosaicing and conversion to this 1.0 pRGB.  It is after Scene Reconstruction and before Creative Processing, to lend these terms from Simon Tindemans’ essay [a href=\"http://21stcenturyshoebox.com/essays/scenereferredworkflow.html]here[/url].

Blending of such a scene-referred image (in 1.0 pRGB) with an already processed “JPEG” from one single shot (and brought into same 1.0 pRGB and at high bit depth) could be seen as a creative tool and part of Creative Processing on the way to an output-referred i.e. preferred rendition. Nothing mandatory just an option.

There are for sure pros and cons with this approach. FWIW, the main advantage I see is to let the specific “look” e.g. from a Canon Picture Style shine through, without a variety of adjustments needed - at least for a start. And while keeping key Raw advantages such as highlight details. The term “shine through” could refer to a normal RGB blend at reduced but tunable Opacity, or, maybe there are better ways e.g. via HSL blend modes.

Another possible advantage concerns very saturated colors. While the pRGB pathway is certainly suited to preserve a major part of scene-referred color saturation, many of them still can not be printed, so that the user often has to go through a second round of editing in order to get such oog colors towards or inside the designated output gamut while keeping the appearance as far as possible. Purists will of course appreciate the flexibility with this approach, but it might not be everyone’s case.
Whereas camera manufacturers’ seem to have their own proprietary ways to compress saturated colors into tiny sRGB while preserving “vividness” and image details, again, as far as possible. At least, for me their recipe doesn’t look like a straight RelCol pathway which tends to produce more channel clipping and posterization.
By means of said Raw + JPG blend in 1.0 pRGB as described above, absolute color saturation would be “buffered”, so that the de facto gamut as occupied by the image can already be closer to the designated output gamut (in an global sense).

To raise a last aspect; aside from suggested “Raw + JPEG” amalgamation, requested capability to blend images in ACR at the level of 1.0 pRGB could also be particularly useful e.g. for HDR imaging from “Raw + Raw” by doing it “properly” in a linear-gamma space and before Creative Processing.

Makes sense, or perhaps not enough?
Let’s see …

Peter

--
« Last Edit: March 27, 2008, 02:59:36 PM by DPL » Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 8929



WWW
« Reply #21 on: March 27, 2008, 03:16:47 PM »
ReplyReply

Quote
Referring to Camera Raw and the processing stage of linear-gamma ProPhotoRGB (1.0 pRGB) which ACR was reported to use as an intermediate working space, there should be access to scene-referred image data, freshly derived from “native” Raw by basic operations such as demosaicing and conversion to this 1.0 pRGB.  It is after Scene Reconstruction and before Creative Processing, to lend these terms from Simon Tindemans’ essay here.

I'm not positive, I'd love to hear from Schewe or Thomas Knoll/Mark Hamburg if indeed these setting really are providing scene referred rendering. I once asked Mark and the answer wasn't all that clear:

Quote
At what point things shift from scene referred to output referred is more or
less in the eye of the beholder. After all, if you set everything flat all
the way through, then there isn't really a transition point.

I can see the need for some, few users who might want scene referred data out of a converter. I'm not sure where they will now end up with output referred (Photoshop one presumes) or why they would do the toning there instead of LR/ACR. I don't find the toolset there "better", its certainly slower going. And its not anywhere as non destructive unless you go way out of your way with lots of adjustment layers. Seems like an interesting exercise but not something I'd do other than just an exercise.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
Peter_DL
Sr. Member
****
Offline Offline

Posts: 421


« Reply #22 on: March 27, 2008, 04:56:37 PM »
ReplyReply

Quote
I'm not positive, I'd love to hear from Schewe or Thomas Knoll/Mark Hamburg if indeed these setting really are providing scene referred rendering. I once asked Mark and the answer wasn't all that clear: At what point things shift from scene referred to output referred is more or less in the eye of the beholder. After all, if you set everything flat all the way through, then there isn't really a transition point. [a href=\"index.php?act=findpost&pid=184750\"][{POST_SNAPBACK}][/a]
Andrew, - you and Mark Hamburg are of course right: there’s no clear distinction. As I understand, white balance would be a good example for such a “hybrid” between accurate scene-referred and pleasing.  However, there’s perhaps an easy way to make a clear distinction, arguable though:

All ACR controls which are applied before or during the conversion Raw -> 1.0 pRGB would count for “Scene Reconstruction” to achieve still somewhat scene-referred image data. All ACR controls which are applied after arrival in 1.0 pRGB would count for “Creative Processing” to be applied on such scene-referred image data.

Quote
I can see the need for some, few users who might want scene referred data out of a converter. I'm not sure where they will now end up with output referred (Photoshop one presumes) or why they would do the toning there instead of LR/ACR. I don't find the toolset there "better", its certainly slower going. And its not anywhere as non destructive unless you go way out of your way with lots of adjustment layers. Seems like an interesting exercise but not something I'd do other than just an exercise.
[a href=\"index.php?act=findpost&pid=184750\"][{POST_SNAPBACK}][/a]
My initial post was/is meant as an ACR feature request, and (as a non-native speaker) I would have hoped that at least the mechanistic description was sufficiently clear, however, I’m sorry for any possible confusion from terminology.

The extraction of such "scene-referred image data" (see I'm putting it in quotation marks now) from ACR was of course an helpful exercise to simulate the feature request. It was not a blind flight.

Peter

--
Logged
madmanchan
Sr. Member
****
Offline Offline

Posts: 2108


« Reply #23 on: March 27, 2008, 06:04:20 PM »
ReplyReply

Quote
I'm not positive, I'd love to hear from Schewe or Thomas Knoll/Mark Hamburg if indeed these setting really are providing scene referred rendering. I once asked Mark and the answer wasn't all that clear:

You can effectively get linear output from ACR by setting Brightness = 0, Contrast = 0, Blacks to zero, and set the Point Curve to Linear. Of course make sure there aren't any other settings applied such as sharpening, noise reduction, parametric curves, HSL tuning, etc.

If you save the output as ProPhoto, the numerical values will of course be gamma-encoded using ProPhoto's gamma of 1.8. So take the resulting numbers (e.g., using a color picker in PS) and compute x ^ 1.8 to return them to ProPhoto linear, and there you have it.

There is not a way from ACR to disable the camera profiles. (Obviously in normal use you'd never want to disable them!) What you can accomplish using the steps described above is get the demosaiced data that is white-balanced and converted to ProPhoto linear, nothing else. (That's not strictly true, but it's mostly true.)

Anyone wishing to learn more about the process is welcome to examine the DNG spec (and for software developers, examine the DNG SDK). As I recall, dng_validate in the DNG SDK can dump the raw data in various stages, including before the camera profile is applied.
Logged

madmanchan
Sr. Member
****
Offline Offline

Posts: 2108


« Reply #24 on: March 27, 2008, 06:09:01 PM »
ReplyReply

I would add that IMO the most significant edit that distinguishes between scene-referred and output-referred data is the tone curve. Without the tone curve we are seeing linear data which mirrors the sensor's response (in terms of relative energy, not color), which is not at all how human vision works.

Color geeks are aware that L* in CIE L*a*b* is more perceptually uniform and know that relationship between L* and Y (in XYZ) is highly non-linear.
Logged

digitaldog
Sr. Member
****
Offline Offline

Posts: 8929



WWW
« Reply #25 on: March 27, 2008, 06:11:00 PM »
ReplyReply

Quote
You can effectively get linear output from ACR by setting Brightness = 0, Contrast = 0, Blacks to zero, and set the Point Curve to Linear. Of course make sure there aren't any other settings applied such as sharpening, noise reduction, parametric curves, HSL tuning, etc.

OK assuming we agree this is scene referred, why? What do you do with it now? And why not use the controls to get output referred?

Quote
There is not a way from ACR to disable the camera profiles

Since the two profiles are used to set a white balance, it seems here alone we're getting some kind of output referred alteration if that's the correct term to use. IOW, even using the defaults, isn't some WB handling happening by virtue of these two profiles you can't 'turn off'?
« Last Edit: March 27, 2008, 06:12:50 PM by digitaldog » Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
madmanchan
Sr. Member
****
Offline Offline

Posts: 2108


« Reply #26 on: March 27, 2008, 09:52:13 PM »
ReplyReply

Quote
OK assuming we agree this is scene referred, why? What do you do with it now? And why not use the controls to get output referred?

Oh, I'm not suggesting that this linear output I described earlier is useful for anything. I was just describing the steps required to produce that output, if for some reason somebody wanted to see it (e.g., out of curiosity).

Certainly to get a pleasing visual result you would want to use controls typically found in RAW converters (such as the tone curve).


Quote
Since the two profiles are used to set a white balance, it seems here alone we're getting some kind of output referred alteration if that's the correct term to use. IOW, even using the defaults, isn't some WB handling happening by virtue of these two profiles you can't 'turn off'?
[a href=\"index.php?act=findpost&pid=184799\"][{POST_SNAPBACK}][/a]

Yes, but I would argue it's still scene-referred colorimetry because each RGB coordinate still has a linear relationship to the non-white-balanced camera coordinates. In other words, the interpolated camera -> XYZ matrix is providing a scene-referred colorimetric interpretation of what those captured camera RGB values mean.

If we were to examine the data before the cam -> XYZ matrices are applied, then we'd certainly be seeing the data in a more "raw" state (less processed) but it wouldn't be very meaningful because there wouldn't be any colorimetry involved. Sort of like examining a printer profile target which is just untagged RGB.
Logged

digitaldog
Sr. Member
****
Offline Offline

Posts: 8929



WWW
« Reply #27 on: March 27, 2008, 10:21:58 PM »
ReplyReply

Quote
Oh, I'm not suggesting that this linear output I described earlier is useful for anything. I was just describing the steps required to produce that output, if for some reason somebody wanted to see it (e.g., out of curiosity).

Certainly to get a pleasing visual result you would want to use controls typically found in RAW converters (such as the tone curve).
Yes, but I would argue it's still scene-referred colorimetry because each RGB coordinate still has a linear relationship to the non-white-balanced camera coordinates. In other words, the interpolated camera -> XYZ matrix is providing a scene-referred colorimetric interpretation of what those captured camera RGB values mean.

OK, all sounds good, thanks. There was a link to an article discussing some "benefits" of setting ACR/LR for scene referred output but the arguments for doing so seemed less than useful to me. I was wondering if I was missing something.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
madmanchan
Sr. Member
****
Offline Offline

Posts: 2108


« Reply #28 on: March 28, 2008, 08:15:54 AM »
ReplyReply

No, I think we're on the same page.

Following the "produce linear output" steps I described above is a nice exercise for the reader in understanding why scene-referred rendering looks nasty and unattractive. It also shows how critical the tone curve is to crafting the image so that its values have relationships that are more in line with how our visual system works.

I am puzzled by why some users would think that getting linear output from ACR or any raw converter would be a better starting point for further edits. That's totally bizarre to me. The only reason I can see for that is if (1) you're doing scientific work and need the linear output for some purpose or (2) you're not happy with the raw converter's built-in set of editing tools.

But for example, if you get linear output out of ACR and bring it into PS, then yes you can implement your own white balance by using, say, the Curves tool to set different clipping points for each of the R, G, and B values. But there's no technical advantage of doing so compared to doing it in the raw converter in the first place (and arguably a disadvantage since placing curve points inside PS is not as precise as using the click-WB tool inside ACR).

Similarly, one can "turn off" the tone curve in ACR and bring the linear output into PS and apply the tone curve instead in PS. But again there's not a technical advantage of doing this. It just takes more work! The only reason I could see for doing this is if you're not happy with ACR's tone curve controls. Given that ACR's tone curve implementation takes some steps to minimize hue shifts compared to PS's RGB curves tool (by default), I would say that for the vast majority of cases it's preferable to shape the tones in ACR ...

Eric
Logged

Pages: « 1 [2]   Top of Page
Print
Jump to:  

Ad
Ad
Ad