Ad
Ad
Ad
Pages: « 1 [2] 3 »   Bottom of Page
Print
Author Topic: "Settings for an Accurate Histogram": WB?  (Read 17245 times)
digitaldog
Sr. Member
****
Offline Offline

Posts: 9225



WWW
« Reply #20 on: September 24, 2009, 03:02:58 PM »
ReplyReply

Quote from: bjanes
Do Canon cameras permit rendering into a wider space than aRGB by the camera?

Adobe RGB yes. But its not much help.
Logged

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

Posts: 87


WWW
« Reply #21 on: September 25, 2009, 03:07:47 PM »
ReplyReply

I've been doing some more experimenting with this now because I thought I had it all figured out before, but hadn't considered saturation clipping and the fact that Lightroom is using Melissa RGB with the camera using Adobe RGB for its preview.

What I've found is that I actually don't want to have contrast all the way down at -4. If I'm using a ‘neutral’ white balance (the RGB multipliers at 1) and go by the clipping indicators on the camera, I end up with bad looking highlights.

I use a custom DNG profile with my camera in Lightroom and have most of the picture controls at zero by default, using a linear tone curve. From my testing, it seems that if I have to go below around -0.22 on the exposure correction (before adjusting white balance) I may be able to recover textural detail, but it's badly posterised/discoloured.

So while there may be more highlight detail there in the RAW file, and -4 contrast might be a better indicator of that, there doesn't seem to be any more useful detail above that point. (note: after changing the white balance, I can go much lower than -0.22, due to the way Lightroom seems to change it)

I don't know if other RAW converters may be able to get more useful detail out of the highlights without ending up discoloured, but that seems to be the limit with my 1000D in Lightroom at least.

To get the highlight clipping indicators on the camera to match that point, the neutral picture style at zero contrast is closest.
I then find that -4 saturation gets the rest of the histogram to be a close match to Lightroom's. (due to it using a wider colour space)

It's not a perfect match, especially very near black (the RAW file has more usable shadow detail than the camera preview) but it is at least useful for knowing where your highlights are going to clip badly.



Maybe if you're setting the white balance in the camera rather than putting it to a neutral position, or shooting in greyscale, -4 contrast is a good idea to help avoid clipping a single channel, but with my camera at least, and when using a neutral white balance, it's actually making me over-expose images.
Logged
bjanes
Sr. Member
****
Online Online

Posts: 2864



« Reply #22 on: September 25, 2009, 03:16:40 PM »
ReplyReply

Quote from: digitaldog
Adobe RGB yes. But its not much help.

Isn't aRGB Adobe RGB?
Logged
bjanes
Sr. Member
****
Online Online

Posts: 2864



« Reply #23 on: September 25, 2009, 03:20:25 PM »
ReplyReply

Quote from: joedecker
Nice article.  One question, though....

For per-channel histograms, on cameras that create a histogram even for RAW images based on the associated JPG, doesn't the white balance setting affect the relative brightnesses of the channels and therefore affect the accuracy of the histogram as well?

Yes it does. That is why UniWB helps--it sets the WB multipliers to 1 so that the resulting histograms which are derived from the JPG give a better indication of the raw channels. Saturation clipping into a small color space such as aRGB or sRGB can still occur.
Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 9225



WWW
« Reply #24 on: September 25, 2009, 03:22:33 PM »
ReplyReply

Quote from: bjanes
Isn't aRGB Adobe RGB?

Presumably it is. I don’t use the “term” aRGB, but I suspect anyone who does means Adobe RGB (1998).
Logged

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

Posts: 1805


« Reply #25 on: September 25, 2009, 05:25:49 PM »
ReplyReply

Quote from: Andrew Fee
What I've found is that I actually don't want to have contrast all the way down at -4. If I'm using a ‘neutral’ white balance (the RGB multipliers at 1) and go by the clipping indicators on the camera, I end up with bad looking highlights
1. The advice to set the contrast to anything non-neutral (i.e. positive or negative) is misguided. Positive contrast increases the highlights and decreases the shadows, thereby causing raw conversion clipping. Negative contrast does the opposite, thereby hiding present raw clipping.

2. It plays no role, how the highlights - or the shadows - look in an ETTRed raw image without appropriate adjustments. The goal of ETTR is to capture as much light as possible without raw clipping (sometimes with a tolerable amount of raw clipping). The "look" of the shot is the question of raw conversion.
Logged

Gabor
bjanes
Sr. Member
****
Online Online

Posts: 2864



« Reply #26 on: September 25, 2009, 05:59:24 PM »
ReplyReply

Quote from: Andrew Fee
I've been doing some more experimenting with this now because I thought I had it all figured out before, but hadn't considered saturation clipping and the fact that Lightroom is using Melissa RGB with the camera using Adobe RGB for its preview.

What I've found is that I actually don't want to have contrast all the way down at -4. If I'm using a ‘neutral’ white balance (the RGB multipliers at 1) and go by the clipping indicators on the camera, I end up with bad looking highlights.

I use a custom DNG profile with my camera in Lightroom and have most of the picture controls at zero by default, using a linear tone curve. From my testing, it seems that if I have to go below around -0.22 on the exposure correction (before adjusting white balance) I may be able to recover textural detail, but it's badly posterised/discoloured.

Andrew, you don't say what camera you are using, but are you aware of the baseline exposure offset that Lightroom and Camera Raw uses. It varies with the camera. For My Nikon D3, the offset is +0.5 EV meaning that highlights that are not blown in the raw file may appear to be blown in ACR/Lightroom unless one takes this offset into account. For this camera, if I set the Exposure in ACR to -0.5 EV and set the other sliders all to zero and the point curve to zero, a so called linear TRC even though it is in the gamma of the working space, I get good correlation between the RGB histograms in Rawnalize and ACR for the green channel, which has no WB multiplier. Saturation clipping in aRGB may occur in some images, but is not that common.
Logged
Andrew Fee
Jr. Member
**
Offline Offline

Posts: 87


WWW
« Reply #27 on: September 25, 2009, 10:36:35 PM »
ReplyReply

Quote from: bjanes
Andrew, you don't say what camera you are using, but are you aware of the baseline exposure offset that Lightroom and Camera Raw uses. It varies with the camera. For My Nikon D3, the offset is +0.5 EV meaning that highlights that are not blown in the raw file may appear to be blown in ACR/Lightroom unless one takes this offset into account. For this camera, if I set the Exposure in ACR to -0.5 EV and set the other sliders all to zero and the point curve to zero, a so called linear TRC even though it is in the gamma of the working space, I get good correlation between the RGB histograms in Rawnalize and ACR for the green channel, which has no WB multiplier. Saturation clipping in aRGB may occur in some images, but is not that common.
It's a Canon 1000D, which has a +0.25 correction, if I remember correctly. When using the custom DNG profile I made with a ColorChecker, -0.22 seems to work better.

Maybe I wasn't too clear with what I was trying to say though. -0.25 exposure is the lowest I can go in Lightroom and get useful data back. I can still go lower and bring back more detail, but it's badly discoloured.

If I set the camera to the neutral preset with 0 contrast, the clipping indicator on the camera matches up nicely with Lightroom at -0.25.

If I set contrast as low as possible (-4) as recommended in this article, the clipping indicator only shows when the RAW file is entirely clipped—the point where I can get nothing back at all, rather than the point where I can't recover any more useful information. This isn't really very helpful, as you're potentially moving information you want to keep into an area where you can recover textural detail, but not colour when exposing to the right.
« Last Edit: September 25, 2009, 10:37:16 PM by Andrew Fee » Logged
Jonathan Wienke
Sr. Member
****
Offline Offline

Posts: 5759



WWW
« Reply #28 on: September 26, 2009, 07:43:46 AM »
ReplyReply

Quote from: Panopeeper
1. The advice to set the contrast to anything non-neutral (i.e. positive or negative) is misguided. Positive contrast increases the highlights and decreases the shadows, thereby causing raw conversion clipping. Negative contrast does the opposite, thereby hiding present raw clipping.

This is not universally true; the meaning and range of JPEG adjustment parameters varies widely, depending on the brand and model of the camera. The only way to know how a particular camera setting affects the relation between RAW and JPEG values is to test it for yourself.

Quote
2. It plays no role, how the highlights - or the shadows - look in an ETTRed raw image without appropriate adjustments. The goal of ETTR is to capture as much light as possible without raw clipping (sometimes with a tolerable amount of raw clipping). The "look" of the shot is the question of raw conversion.

I agree with this, and find it amazing how many photographers, even supposedly master-level photographers, who haven't figured this out yet. It's the root of much of the "MFDB vs. DSLR" debate.
Logged

Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1325



WWW
« Reply #29 on: September 26, 2009, 09:37:12 AM »
ReplyReply

Quote from: Jonathan Wienke
I agree with this, and find it amazing how many photographers, even supposedly master-level photographers, who haven't figured this out yet. It's the root of much of the "MFDB vs. DSLR" debate.
Film culture I think. Those master-level photographers never faced a media with such linear properties as a digital sensor, and now tend to extrapolate their past know how. I am tired of hearing people talking about changes in hue and saturation depending on digital exposure.

Regards
Logged

bjanes
Sr. Member
****
Online Online

Posts: 2864



« Reply #30 on: September 30, 2009, 08:24:52 AM »
ReplyReply

Quote from: Panopeeper
1. The advice to set the contrast to anything non-neutral (i.e. positive or negative) is misguided. Positive contrast increases the highlights and decreases the shadows, thereby causing raw conversion clipping. Negative contrast does the opposite, thereby hiding present raw clipping.

There is a great deal of confusion regarding positive and negative contrast settings for the camera JPEG rendering. On the Nikon D3, normal contrast (contrast = 0 in the picture control) applies a rather strong S curve to the raw data, lightening the quarter tones and darkening the three quarter tones. A contrast setting of -3 applies a slightly less strong S curve to the data, but it is still a positive contrast curve where the light quarter tones are lightened and the dark quarter tones are further darkened. The highlight clipping point is not affected by a contrast curve, since input equals output at an 8 bit value of 255. If the camera histogram is conservative and indicates clipping in the JPEG conversion when there is none in the raw file, then a lower contrast curve will lower the quarter tones and decrease clipping in the histogram.

Shown below are TRCs for the Nikon D3 with normal contrast (picture control contrast = 0) on the top and reduced contrast (picture control contrast = -3) on the bottom. These TRCs were produced from the same raw file by varying the picture control settings in CaptureNX, which would closely approximate the in camera settings.

[attachment=16896:NxContrasts.gif]

The effects of curves can be shown in ACR. This Stouffer wedge demonstrates clipping in Step 1 with a linear tone curve.
[attachment=16897:LinearCurve.png]

The clipping is unchanged with a positive contrast curve. The quarter tones are raised as shown. The value of 194 in the file is raised to 251. If the camera histogram were conservative and shows clipping at 251, then the histogram would show clipping when there is none.
[attachment=16898:PositiveContrast.png]

A true negative contrast curve will decrease the quarter tones as shown below. A value of 193 is reduced to 176. A conservative camera histogram would show less clipping.
[attachment=16899:NegContrast.png]

However, true clipping can not be eliminated by a contrast curve, since at the right end of the curve, input is equal to output. To eliminate the clipping, the curve has to decrease the highlights slightly as shown below. If the camera histogram is showing clipping when there is none, the best solution would be to upload a custom curve to the camera that decreases the highlights.
[attachment=16900:DecreaseHighs.png]




Logged
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #31 on: October 03, 2009, 02:51:48 PM »
ReplyReply

Quote from: bjanes
There is a great deal of confusion regarding positive and negative contrast settings for the camera JPEG rendering. On the Nikon D3, normal contrast (contrast = 0 in the picture control) applies a rather strong S curve to the raw data, lightening the quarter tones and darkening the three quarter tones. A contrast setting of -3 applies a slightly less strong S curve to the data, but it is still a positive contrast curve where the light quarter tones are lightened and the dark quarter tones are further darkened
I guess the curve can be loaded with the D3 as well. Thus the effects of the "contrast" setting and the curve should be separated. One should load a linear curve, or even better, an inverse sRGB (or Adobe RGB) curve, to balance the non-linear mapping associated with the color space.

Quote
The highlight clipping point is not affected by a contrast curve, since input equals output at an 8 bit value of 255
I suggest you to test it with any raw file in ACR:

1. reset contrast to 0,
2. adjust the exposure so, that it is close to but under clipping,
3. increase the contrast; clipping will occur,
4. reset the contrast again and increase the exposure, so that some clipping occurs,
5. decrease the contrast (i.e. go into negative contrast): the clipping will disappear.

I don't think the in-camera JPEG creation differs from the above.
Logged

Gabor
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #32 on: October 03, 2009, 08:47:50 PM »
ReplyReply

I recall experimenting with my 5D in Hanoi several years ago, shooting the same scene out of the hotel window and changing the contrast settings of the Landscape Picture Style for each shot. The minimum contrast setting seemed to produce the most accurate histogram with regard to ETTR. At least it corresponded most accurately to the ETTR rendition of ACR after an appropriate amount of negative EC.

However, I have since experienced feedback from Gabor, with regard to some of my images I've sent him, that what I've experienced in ACR as an accurate ETTR, is in fact slightly clipped in the extreme highlights according to his Rawnalyze program.

The issue then becomes, 'does it really matter to the practising photographer if an image appears to be correctly exposed to the right when by certain technical standards it's actually very slightly overexposed?'

I remember on the same occasion in Hanoi experimenting with another method of achieving a good ETTR. Set camera to spot meter mode. Focus on the brightest part of the image, whether a bright cloud or a white napkin in a restaurant. Take note of the exposure reading, then increase exposure by 3 stops. You should get a good ETTR with the 5D without even looking at the histogram.

Logged
bjanes
Sr. Member
****
Online Online

Posts: 2864



« Reply #33 on: October 05, 2009, 06:19:16 PM »
ReplyReply

Quote from: Panopeeper
I guess the curve can be loaded with the D3 as well. Thus the effects of the "contrast" setting and the curve should be separated. One should load a linear curve, or even better, an inverse sRGB (or Adobe RGB) curve, to balance the non-linear mapping associated with the color space.


I suggest you to test it with any raw file in ACR:

1. reset contrast to 0,
2. adjust the exposure so, that it is close to but under clipping,
3. increase the contrast; clipping will occur,
4. reset the contrast again and increase the exposure, so that some clipping occurs,
5. decrease the contrast (i.e. go into negative contrast): the clipping will disappear.

I don't think the in-camera JPEG creation differs from the above.

Gabor,

As requested, more or less. Here is ACR with a linear TRC, contrast = 0, minimal clipping in the highlights as shown in red. I am rendering into a 16 bit space.
[attachment=16974:SlClip_contrast0.png]

Increasing the contrast all the way to 98 causes only a very minimal difference:
[attachment=16975:SlClip_contrast98.png]

My original assertion stands. Why would contrast affect clipping when a curve does not affect values at the extreme right end of the curve? The situation is analogous to clipping with the brightness curve. As Jeff Schewe states in his ACR book, setting brightness to 100 looks like clipping, but if you check the 16 bit file you will find that the values have been pushed to max, but there is no clipping. The same applies to contrast.

Regards,

Bill
Logged
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #34 on: October 05, 2009, 09:27:22 PM »
ReplyReply

Quote from: bjanes
As requested, more or less. Here is ACR with a linear TRC, contrast = 0, minimal clipping in the highlights as shown in red. I am rendering into a 16 bit space.
Bill,

you almost caused me a heart attack. Before I posted my previous message, I loaded a raw image in ACR without any particular selection and did just what I suggested, and the effect was just what I described. Now, I wanted to repeat that, and I picked a raw file - and it did not work. Then I pickd another one - it did not work. I tried about six raw images, until I found one, which reacted just like I described.

The first capture is with exposure +0.90; this is the maximum intenity increase without clipping; the contrast is 0 (tone curve: linear in all samples):

Exposure +0.9 Contrast 0

Now the same exposure, but contrast +100; it obviously created lots of clipping:

Exposure +0.9 Contrast 100

Next, the intensity increased by 1.1; this causes clipping. The contrast is 0:

Exposure +1.1 Contrast 0

Finally, the same intensity with contrast -50, and the clipping is gone:

Exposure +1.1 Contrast -50

This effect is natural to me; particularly, decreasing the contrast must pull the higher intensity pixels back, i.e. away from clipping. Now, I'd like to know why this is not working that way with all images...
Logged

Gabor
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #35 on: October 05, 2009, 09:45:07 PM »
ReplyReply

Quote from: bjanes
My original assertion stands. Why would contrast affect clipping when a curve does not affect values at the extreme right end of the curve? The situation is analogous to clipping with the brightness curve. As Jeff Schewe states in his ACR book, setting brightness to 100 looks like clipping, but if you check the 16 bit file you will find that the values have been pushed to max, but there is no clipping. The same applies to contrast.


I'm getting a bit confused here, Bill. It's always possible to apply curves which increase contrast in certain parts of the image but not others. The standard 'brightness/contrast' tool in earlier versions of Photoshop would very easily blow highlights. In CS3 and beyond, those highlights are protected and the tool is now much more useful.

The issue of getting the camera's histogram to accurately reflect exposure for a full ETTR has to do with that flashing highlight-warning, does it not?

With the Canon 5D, setting contrast to zero (in one of the picture Styles) results in the histogram providing misleading information in respect to a degree of highlight-warning flashing. As a consequence, one tends to reduce exposure when it may not be necessary. Setting the contrast to -4 (on the 5D) fixes this problem. It's as simple as that, isn't it?
Logged
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #36 on: October 05, 2009, 09:57:25 PM »
ReplyReply

Quote from: Ray
With the Canon 5D, setting contrast to zero (in one of the picture Styles) results in the histogram providing misleading information in respect to a degree of highlight-warning flashing
There is no point in trying to coax the camera into displaying raw-like histograms with only a subset of the related adjustments. If the saturation is not 0,  if the white balance is not neutral (i.e. the coefficients are not close to 1.0), then the contrast setting will not make up for all those. A certain contrast setting may "repair" the histogram in one setting, but not in another, particularly because of the changing WB.

(The tonal curve changes the appearance of the histograms, but it does not cause or eliminate clipping.)

Quote
It's as simple as that, isn't it?
Well, good for you.
Logged

Gabor
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #37 on: October 05, 2009, 10:56:56 PM »
ReplyReply

Quote from: Panopeeper
There is no point in trying to coax the camera into displaying raw-like histograms with only a subset of the related adjustments. If the saturation is not 0,  if the white balance is not neutral (i.e. the coefficients are not close to 1.0), then the contrast setting will not make up for all those. A certain contrast setting may "repair" the histogram in one setting, but not in another, particularly because of the changing WB.

(The tonal curve changes the appearance of the histograms, but it does not cause or eliminate clipping.)


Well, good for you.
 
Gabor,
I take it you are referring to the one setting in one particular 'Picture Style' not being suitable for all situations. This is true. After setting the contast to minimum in the 'Lanscape' picture style on the 5D, the histogram tends to be accurate only for landscape-type shots. It's not accurate for shots in a dimly lit theatre using flash, for example. I haven't experimented with all combinations.

Quote
Well, good for you.

Thank you. I like to keep it simple.
Logged
bjanes
Sr. Member
****
Online Online

Posts: 2864



« Reply #38 on: October 06, 2009, 06:46:00 AM »
ReplyReply

Quote from: Ray
With the Canon 5D, setting contrast to zero (in one of the picture Styles) results in the histogram providing misleading information in respect to a degree of highlight-warning flashing. As a consequence, one tends to reduce exposure when it may not be necessary. Setting the contrast to -4 (on the 5D) fixes this problem. It's as simple as that, isn't it?

Ray, I think that it is not that simple. My thesis is that a contrast curve affects the quarter tones and has no effect at highlight saturation where input equals output. If the camera warning is conservative and indicates clipping when there is none in the raw file, then lowering the contrast will decrease those highlights in the quarter tones that were causing the clipping. However, if the highlight warning is perfectly accurate, then contrast will have no effect.
Logged
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #39 on: October 06, 2009, 06:23:37 PM »
ReplyReply

Quote from: bjanes
My thesis is that a contrast curve affects the quarter tones and has no effect at highlight saturation where input equals output. If the camera warning is conservative and indicates clipping when there is none in the raw file, then lowering the contrast will decrease those highlights in the quarter tones that were causing the clipping. However, if the highlight warning is perfectly accurate, then contrast will have no effect.
I do realize, that different cameras may apply contrast differently, although I have a hard time to believe, that a negative setting causes positive contrast adjustment.

I made a test again with my camera, Canon 40D, and I found, that the flashing clipping indication as well as the appearance of the histogram does depend on the contrast setting. Negative contrast "hides" clipping, while positive contrast increases it. I don't believe, that the accuracy of highlight warning has to do with the way the contrast adjustment works. This accuracy is our special subject, the camera's firmware does not give a damn for it, IMO.
Logged

Gabor
Pages: « 1 [2] 3 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad