Pages: [1]   Go Down

Author Topic: Shouldn’t have done this test: Exposure vs. DNG Profiles from Chart Wizard  (Read 12143 times)

Peter_DL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 544

 
It is about Exposure vs. DNG Profiles from Chart Wizard of the DNG Profile Editor.

Two shots, +/- 1 EV apart, of the ColorChecker Chart taken in quick succession at somewhat normal sunlight (no controlled lab environment – admittedly). The Raw files were DNGed and supplied to the DNG Profile Editor.

After running the ChartWizard, and successful built of the Color table, the finding is that some of the color adjustments per color patch deviate considerably between both shots. See example below which for the first skin tone patch gives Hue +2, Saturation -8 for the darker capture versus Hue -2, Saturation -2 for the brighter shot.

Is this an expected deviation ?
Or, are we in need of lightness/exposure-dependent Hue/Sat.-corrections,
considering that the Exposure slider in ACR was stated to be applied after the Hue/Sat.-table ?

Peter

--
Logged

b2martin

  • Full Member
  • ***
  • Offline Offline
  • Posts: 136

Could this be due to clipping in the +1EV shot?
Logged

Peter_DL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 544

Could this be due to clipping in the +1EV shot?

The camera is not supported by Rawnalyze.
No indication for clipping with the +1EV shot in ACR though (see below). No complaints from the DNG PE Chart Wizard.
The patches being most subject to different Hue/Sat.-adjustments depending on +/-EV real Exposure are not the brightest ones.

Thanks for your thoughts.

Peter

--
Logged

b2martin

  • Full Member
  • ***
  • Offline Offline
  • Posts: 136

Have you looked at the RGB values of all the squares to see if any are at 255?
Logged

Peter_DL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 544

Many thanks for bringing this up again.

I'm nearly 100% sure that there are no Raw clipping issues involved with the above example,
however, I'll be happy to repeat the test with a rawnalyzable Camera
- once it stops raining here in Central Europe.

Peter

--
Logged

Tim Lookingbill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2436

I would suspect the reasons for your findings illustrates quite clearly that digital sensors aren't entirely linear. Also the software interpretation of the sensor data should be suspect as well.

Every time I lighten an image in Adobe Camera Raw whether I use a custom DNG profile or not, I constantly get  hue errors mainly introduced in combination with the appearance of warm/cool WB and the erratic, non-gradual change to contrast ratio with every increase of luminance. The lighter the colors get usually requires I flatten a section of the default Medium S-Curve to correct for the hue shift.
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 20630
  • Andrew Rodney
    • http://www.digitaldog.net/

Every time I lighten an image in Adobe Camera Raw whether I use a custom DNG profile or not, I constantly get  hue errors mainly introduced in combination with the appearance of warm/cool WB and the erratic, non-gradual change to contrast ratio with every increase of luminance. The lighter the colors get usually requires I flatten a section of the default Medium S-Curve to correct for the hue shift.

And yet, in this post we hear:

Quote
Hue twists are frequently attributed to ACR, but Eric Chan has stated in another thread that if one uses ETTR and normalization of the image is necessary, one should use the exposure slider first to get the tonal values in the proper range before the 3D color table is applied.
http://www.luminous-landscape.com/forum/index.php?topic=56863.40

Are you reporting something different here?
Logged
http://www.digitaldog.net/
Author "Color Management for Photographers".

Peter_DL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 544

Are you reporting something different here?

Yes.
Sincerely, Peter

--
Logged

Tim Lookingbill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2436

Since this subject has been brought up, I'ld like to get other's thoughts on the cause of similar behavior adjusting exposure with or without the use of a custom profile.

There seems to be a need (at least for myself) for a luminance slider that brightens an image equally and smoothly down the tone scale without color distortions, not only those shown in Peter's +/1 EV adjust CC chart samples, but also to prevent the flattening of modeling detail in textures such as rocks, feathers or clouds the lighter this detail becomes with each Exposure slider increase. See the samples below.

The only adjustments I made from ACR 4.6 defaults in the image on the left were WB and Black point set to zero to avoid black clipping.

The only adjustment made on the right image was an increase of +.80 Exposure. The Camera profile loaded was the default ACR 4.4. Switching to both a dual and single illuminant custom DNG profile didn't make any difference except shifting the hue to a reddish orange instead of the yellowish orange.

Note the ratio of shadows not lightening with the same degree as the lighter tones. I want to retain the contrast relationship while lightening the image using one slider adjust. I have to always resort to the Fill slider and/or knocking back the Contrast slider which overly flattens everything where I then have to apply heavy tweaks to curves.

Why does this have to involve so many tools just to lighten the image to avoid these distortions? When I go from the shade into broad daylight my shadows don't remain dark while the rest of my body lightens. All tones equally lighten at the same degree while maintaining the same contrast ratio.

Logged

Peter_DL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 544

Also the software interpretation of the sensor data should be suspect as well.

Yes, Tim,

At some point of processing the Raw data need to be colorimetrically interpreted, which is the job of the camera profile i.e. DNG profile from the Chart Wizard,
and it was probably naive (from me) to believe that some 2D Hue/Sat.- corrections (relative to an ideal matrix space) would always be enough without really considering the third dimension of lightness/exposure.

In other words, the idea of + Exposure = linear down-scaling of the Raw RGB data may sound compelling,
but it should not be expected to happen in a perfect "linear" matrix space.

Peter

-
Logged

Tim Lookingbill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2436

How's this for a new adjustment slider in ACR?...

"Perceptual Luminance" right next to the Brightness or Exposure slider. This would maintain both global and local contrast relationships up through the entire tonal scale without creating hue shifts while attempting to lighten the image.

I bet it'll probably make a program engineer go bonkers trying come up with a decent algorithm that'll work the same on all Raw images.
Logged

Peter_DL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 544

 
Old thread. Just to report on some further conclusions here,
after some more tests with bracketed exposures of the ColorChecker.
The question was about Exposure-dependency of the DNG profiles created with the Chart wizard of the DNG Profile Editor.

/> The Hue/Sat.-corrections from the Chart Wizard are indeed found to be dependent on +/- EV camera exposure.
/> The effect is less pronounced with the Hue shifts, but the Saturation shifts vary significantly.
/> Surprisingly, it was found to be a systematic deviation:  the lower the camera exposure, the more are the Sat.-shifts pulled down to lower or negative numbers.  With a dark target capture, it seems that the Chart wizard tries to brighten the patches by reducing saturation. Could be an issue with the underlying HSL separation, rather than sensor imperfections calling for Lightness-dependent Hue/Sat.-corrections (what I initially thought).

I think a most correct exposure for the purpose of running the chart wizard should yield something around L* 96 for the white patch corresponding to the synthetic target in Lab.  The value can be measured directly in the DNG Profile Editor while the readouts are computed based on scene-referred ProPhoto RGB linear coordinates.
L* 96 for the white patch may in practice require a +EV camera exposure setting (depending on camera of course). It is basically in line with Eric’s recent recommendation as quoted below.

Peter

--
The DNG PE needs to be somewhat conservative in its chart wizard with regards to clipping because color tables are always applied after white balance.  Your raw capture may not have clipped data in its native coordinate system (native RGB, without WB applied), but that data can become clipped after WB is applied.  That leads to problems with the color mapping.  DNG PE will generally detect this case and prevent you from proceeding.

My recommendation is simply to bracket exposures and then pick the brightest one that DNG PE will accept without giving you an error.
Logged

Tim Lookingbill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2436

Quote
My recommendation is simply to bracket exposures and then pick the brightest one that DNG PE will accept without giving you an error.

That's the first time I've heard that advice and I've been using the DNG PE since it was made available. Thanks for posting that, Peter.

Wonder what differences there would be in two profiles, one made from the brightest one possible that didn't yield an error compared to one created from a less brighter version.
Logged

samueljohnchia

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 498

Peter, this is very interesting information. I too have been making small investigations regarding the DNG PE, photographic the color checker and dng profiles. Do you find that the saturation values calculated by the DNG PE on a chart that was ETTR (the white patch around L* 96) more closely match the saturation levels of the synthetic color checker? I would prepare my own files for a similar experiment and report my findings.
Logged

samueljohnchia

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 498

That's the first time I've heard that advice and I've been using the DNG PE since it was made available. Thanks for posting that, Peter.

Another reason for choosing a source capture that received as much exposure as possible without total saturation of any color channel is to minimize shot noise, and perhaps drown out pattern noise. This would result in more accurate calculations by the DNG PE, less affected by random noise. For similar reasons one should also avoid shooting the ColorChecker target at ISO 25,600 or the like.
Logged
Pages: [1]   Go Up