Exposing To The Right

[sandymc]

Tone controls such as black point (especially), brightness, and contrast work on individual RGB channels so will cause hue shifts.  I would only expect the exposure slider to behave as a true gain in the resulting output with black point set to zero, brightness and contrast neutral, and the tone curve set to linear.

Well, yes, that's what I thought, and in the ETTR work I did a few years ago, I put the blame for the shifts on a combination of profile and the various controls acting in a non-linear way. But Eric seemed to be saying that ACR will be linear with those controls assuming no saturation

Sandy

[Guillermo Luijk]

I put the blame for the shifts on a combination of profile and the various controls acting in a non-linear way. But Eric seemed to be saying that ACR will be linear with those controls assuming no saturation

Any transformation that means a change in the {R,G,B} relative ratios, means a change in hue/sat, but this is assumed. For instance a simple Bright curve to increase luminance without clipping, will compress the highlights producing desaturation. Even if you work in Lab, and apply the curve over the Luminosity channel, you are changing hue/sat. Lab tries to do it in a perceptually pleasant way, but hue/sat changes happen.

A change in exposure (linear scaling of all three channels) would be the only genuinely linear transformation, keeping the proportions of the RGB values and hence keeping hue/sat. The problem is that pushing exposure will quickly clip some channel so this cannot be the only tool for pp.

I am being simplistic here (for instance with a non standard gamma the above is not 100% true), but basically this is the way exposure works.

Regards

[hjulenissen]

Would the solution be for raw converters to separate global tonal processing into two (clearly labeled) sections, one native camera space, and one standardized perceptually relevant space?

Perhaps a raw file pre-processor that multiplies raw sensel values with some value (assuming that clipping is not an issue) before importing it into a standard, nicely organized raw developer (I guess dcraw could easily be hacked into something like that).

-h

[sandymc]

Guillermo,

Yes, I'm in agreement with you, but that doesn't seem to be what Eric is saying. A few years ago, I said that correcting for ETTR will cause hue shifts, and specifically that such hue shifts occur in LR/ACR. Eric seemed to be saying above that hue shifts won't occur if you adjust the "basic" exposure controls.

Regards,

Sandy

[Peter_DL]

With below image*, and a measuring point set to the non-clipped part of the sky,
minus-ACR-Exposure clearly goes along with a reduction of color saturation.
Hence, there are different scaling factors applied per RGB channel…

* which is a "linear" rendition with relevant all sliders and settings in ACR zeroed out.
Also the Baseline Matrix profile was used in order to avoid any non-linearity from the profile.

Among the various discussions on this subject (see below links), best explanation I've heard is as follows:

… So if you assume clipped white values are infinite, in order to give a pleasing appearance you basically have to use a curve or non-linear exposure adjustment. This is essentially what Adobe does with ACR and Lightroom exposure. The exposure slider isn't really a true "exposure" adjustment equally applied to all values at least when using negative values.

Peter

--


http://www.luminous-landscape.com/forum/index.php?topic=50915.0
http://www.luminous-landscape.com/forum/index.php?topic=49866.msg413890#msg413890
http://forums.adobe.com/message/1210302#1210302

 

[fdisilvestro]

Perhaps a raw file pre-processor that multiplies raw sensel values with some value (assuming that clipping is not an issue) before importing it into a standard, nicely organized raw developer (I guess dcraw could easily be hacked into something like that).

-h

That is exactly the way RawTherapee has implemented it.

[Guillermo Luijk]

that doesn't seem to be what Eric is saying. A few years ago, I said that correcting for ETTR will cause hue shifts, and specifically that such hue shifts occur in LR/ACR. Eric seemed to be saying above that hue shifts won't occur if you adjust the "basic" exposure controls.

Changing exposure won't change hue/sat, but any other transformation will. If correcting exposure is done in the first step (and this is the correct procedure), any later non-linear alteration of hue/sat will be the same, no matter how much the original RAW file was exposed, i.e. no matter if ETTR was used or not.

I think this is what Eric meant, that the ACR/LR exposure slider (the one and only you need for ETTR exposure correction) doesn't produce hue shifts. But Eric is the one who has to speak about what Eric said  

With below image*, and a measuring point set to the non-clipped part of the sky, minus-ACR-Exposure clearly goes along with a reduction of color saturation.
Hence, there are different scaling factors applied per RGB channel…

Peter, are you sure your RAW file had no clipping in any channel? if it had, the test is irrelevant. Of course non-linear tricks happen in the highlights when the minus exposure slider is activated in RAW clipping situations. Basically this (look at the minus-EV curves turning in the last minute to reach (255,255)):



But that is the cleverest strategy, otherwhise clipped highlights would turn gray, and they wouldn't have a correct colour anyway.

Regards

[sandymc]

the one and only you need for ETTR exposure correction

Ahh well, not in the case in ACR - even if every other control is zero'd, that doesn't quite work. Don't know why.

Sandy

[Peter_DL]

 
Guillermo,

With the measuring point placed as shown above, there is no indication for channel clipping in ACR. I mean the (ProPhoto) RGB readings in ACR @ Exposure zero are 236, 217, 141. Also, the linear rendition in DPP shows no clipping at this point in the mid of the sky. Otherwise I don’t have the tools to verify this.

Further, the curved roll-off with negative Exposure as shown in your graph could only explain an increase of saturation, not a reduction.

So I’m clueless here about the non-linearity and loss of saturation observed with -Exposure.

Regards,
Peter

--

[Guillermo Luijk]

With the measuring point placed as shown above, there is no indication for channel clipping in ACR.

Clipping indication in ACR is misleading, ACR is a RAW developer, not a RAW analyzer. So the RAW file could be clipped and ACR will display no clipping (a couple of days ago Ray uploaded a clipped RAW file that ACR with minus exposure said it was OK). If you open your RAW file in Rawnalyze you'll see the actual RAW histograms.

It is not fair assuming ETTR or exposure correction produce hue shifts, when users are not doing proper ETTR but clipping their RAW files. Didn't you find suspicious that the reduction of saturation only happened in the sky?

[Peter_DL]

If you open your RAW file in Rawnalyze you'll see the actual RAW histograms.

I don't think that DPP does any recovery,
however, Rawnalize does not seem to be available online anymore since the author passed away.

Peter

--

[Guillermo Luijk]

I don't think that DPP does any recovery,
however, Rawnalize does not seem to be available online anymore since the author passed away.

If you send me your email I can send it to you. Gabor did a great job, it's a hiper compact .exe that doesn't need any installation since it strictly uses OS libraries. Just double click and begin opening RAW files.

[francois]

I don't think that DPP does any recovery,
however, Rawnalize does not seem to be available online anymore since the author passed away.

Peter

--

You might be able to download it from this site:
http://dave-anderson-photo.com/blog/2010/08/23/gabor-rawnalyze-author-rip/

[Peter_DL]

Many thanks, francois and Guillermo.

IF I'm using the tool correct (?),
there does not seem to be Raw clipping in the mid part of the sky,
where I had originally placed the measuring point.

Peter

--

[madmanchan]

Thanks Bill. That must be the reason for the non-linearity in the calculated curves (which of course doesn't mean any non-linearity in the way Exposure works in ACR).

So if the gamma slope in Adobe RGB's implementation is limited (i.e. Adobe RGB's gamma lifts the shadows less than a pure gamma), and we open an image encoded with a pure 2.2 gamma assigning it to Adobe RGB, we will be seeing the deep shadows a bit brighter than expected. My concern about this is because I use pure 2.2 gamma images and then assign them in Adobe RGB, but I find OK to have some shadow lifting.

Regards

Yes, Guillermo and Bill -- ACR uses a slope-limited initial segment for the deepest shadows.  We do this for both ProPhoto RGB (normally gamma 1. and Adobe RGB (normally gamma 2.2).  You can see the exact math we're using in the public DNG SDK, in case you are curious.  The relevant source file is named dng_color_space.cpp.

This accounts for the shadow discrepancies that you are seeing.

[madmanchan]

So in the minus direction it is really a jump function ?

Clipped white is fixed at RGB 255 while a multiplier < 1 is applied all other values, without any smoothing ?

Hmm

Peter

--

Peter, no it is not a jump function.  It is a smooth function that maps pure white to the rest of the curve below it.  (If it was a step/jump function you would see horrible artifacts in your highlights and upper midtones!)

[madmanchan]

Hi Sandy, thanks for running this test and reporting the results. Emil is right, in that black subtraction (in ACR's case, via Blacks slider) is typically done per-channel so it is expected to result in color changes (including hue). Setting Blacks to 0 will turn off that behavior. Generally speaking, you'll find that a given step/error size will manifest much more strongly in blue than the other colors, because the internal working space of ACR uses the ProPhoto RGB primaries, and the blue primary is pretty far out there ...

[Peter_DL]

Peter, no it is not a jump function.  It is a smooth function that maps pure white to the rest of the curve below it.  (If it was a step/jump function you would see horrible artifacts in your highlights and upper midtones!)

Eric, many thanks for your comment.

In the example posted above an initial RGB triplet of 236, 217, 141 is mapped to 160, 147, 107 by setting the Exposure slider to -0.75 (see screenshots with post # 44). So the multiplier on the R channel is 160/236= 0.678. The multiplier on the B channel is 107/141= 0.759, which is higher. Hence, the channels are squeezed closer together in terms of R:B, and a corresponding loss of saturation is clearly visible as well. Unlike what I initially thought, this cannot be explained by the smoothened roll-off with negative Exposure.

Again, for this example all other tonal controls were zeroed in ACR, which is supposed to provide a "linear rendition". Point curve linear as well. ProPhoto RGB selected for output, which to my knowledge is what the RGB readings refer to. Further, the baseline matrix profile was used to avoid any related hue twists. Also, Rawnalyze does not seem to provide any indication for Raw clipping in this mid part of the sky.

So I’m still puzzled about this effect.

Peter

--

[sandymc]

Hi Sandy, thanks for running this test and reporting the results. Emil is right, in that black subtraction (in ACR's case, via Blacks slider) is typically done per-channel so it is expected to result in color changes (including hue). Setting Blacks to 0 will turn off that behavior. Generally speaking, you'll find that a given step/error size will manifest much more strongly in blue than the other colors, because the internal working space of ACR uses the ProPhoto RGB primaries, and the blue primary is pretty far out there ...

Eric,

Thanks for coming back on this. I ran the same checks with black = 0, contrast and brightness at default. Result is 71.5,82.5,34.2 versus 71.7,83.2,31.6.

So the blue deviation is less, but not eliminated, with black set 0. So far as I can tell, it seems to be also related to the brightness control. I mean, it's not a huge deviation in the grand scheme of things, but still.....

Regards,

Sandy

[Guillermo Luijk]

Rawnalyze does not seem to provide any indication for Raw clipping in this mid part of the sky.

Perhaps ACR's highlight strategy extended beyond the strictly RAW-clipped area. In this thread you can see an example.

I think it makes sense to produce a soft and progressive transition between the OK areas and those where some kind of 'recovery' is needed.

Regards