why does color saturation change with exposure?

[Eric Hiss]

Okay I have always heard that with digital that expose to the right then correct in RAW conversion is the proper way to manage digital capture, however when this is done the resulting images may have nice shadow detail but are less vibrant. If you expose to the left (but not to the point of clipping the shadows) then push up the image exposure in RAW then you get a much more vibrant image.

What's happening - shouldn't the two paths result in the same image? I'm just wondering why they are not. Anyone know what's going on? I've seen this now with both Flexcolor and lightroom RAW conversion tools. Btw - I am working with Leica DMR files on images that had plenty of DR to spare on both ends. Could have exposed +- 2 stops either way and still captured all the data without clipping.

[Bart_van_der_Wolf]

What's happening - shouldn't the two paths result in the same image? I'm just wondering why they are not.

Depending on the Raw processing, there should be no difference other than lower noise. When you pull the exposure in Raw processing after exposing to the right (ETTR) in the camera, there should be no influence on saturation, but there apparently can be a minor shift in color accuracy (I noticed it in ACR, have yet to check C1).

Don't know about Flexcolor, but with Lightroom/ACR4 one has to be careful with the automatic settings. IMHO there is too much highlight recovery instead of exposure reduction with ETTR. The HL recovery capability is huge when needed, but at the expense of flatter highlights (lower contrast).

Bart

[Nick Rains]

Underexposing and correcting "up" will result in a vastly inferior image compared to exposing to the right and then reducing the exposure.

Remember, the 'top' 1 stop contains as many tonal steps as the rest of the image put together. Check this out:

http://www.luminous-landscape.com/tutorials/expose-right.shtml

Also, if using Adobe Camera RAW it's better to use Brightness reduction as opposed to Exposure reduction. You will end up with a snappier image if you do because you are 'stretching' the tones down the range keeping the highlights pegged rather than 'sliding' the whole lot down and thus darkening the highlights.

HTH

[Bart_van_der_Wolf]

Also, if using Adobe Camera RAW it's better to use Brightness reduction as opposed to Exposure reduction. You will end up with a snappier image if you do because you are 'stretching' the tones down the range keeping the highlights pegged rather than 'sliding' the whole lot down and thus darkening the highlights.

While in general that's correct, Eric was specifically referring to the following scenario (unless I misunderstood him):
- The scene contrast (e.g. front-lit) is lower than the camera's dynamic range (with room to spare at the over-exposure end),
- a normal exposure for e.g. mid tones will land those tones in the middle of the range (there are no highlight tones 'at the right' yet),
- ETTR will not overexpose those highlight tones, but the whole scene becomes relatively too bright,
- Saturation changes when pulling general exposure in Raw conversion.

The Saturation/Color shift should not happen.

Bart

[Ray West]

Hi Nick,

Thanks for your input - a very useful link, which explains part of the fun I was having yesterday.

I think a number of the Adobe actions are sort of skewed. Best to test on something simple. If you make an image in ps, just write some coloured text, say. If you copy the image to a number of layers, then zoom in way beyond pixel size. you can see what happens with changes in contrast, gamma, sharpening, etc. This explains why so many sharpened images have the white edges, etc.

This does not take into account the idiosyncrasies of various raw conversions, but does show how easy it is to screw things up afterwards.

Best wishes,

Ray

[Ray West]

Hi Bart,

I think Nick's link sort of explains. The sensors are non-linear in their response. To get both exposures to look the same, you would need to put an exact matching non-linear curve into the conversion. I don't think Eric's conversion is doing that, so it could look more vibrant one way, flatter t'other.

Best wishes,

Ray

[Asher Kelman]

I wonder whether working in an RGB color space has something to do with this. Would it be that LAB or CMYK might protect against this and disconnect brightness from saturation.

Asher

[Bart_van_der_Wolf]

Hi Bart,

I think Nick's link sort of explains. The sensors are non-linear in their response.

Hi Ray,

Actually, sorry to disagree, sensors are not non-linear (as mentioned at the bottom of the referenced LL article). The sensor response is almost perfectly linear to incident light, and the Raw data usually has an equally linear gamma. The Leica M8 'Raw' data is a known exception to the rule. The human eye is far from linear, it more resembles a logarithmic or a power function.

Therefore, there is little reason to expect a color/saturation shift with a uniform scaling operation (exposure slider) before other tonescaling, unless this is a side effect of Adobe's ACR color-management approach (which is initially in Raw more matrix based than LUT profile based).

Bart

[Andrew Rodney]

I wonder whether working in an RGB color space has something to do with this. Would it be that LAB or CMYK might protect against this and disconnect brightness from saturation.

Asher

Nope. There's only one reason to go into CMYK IMHO, to print to a CMYK output device.

Lab, some interesting tricks, most of which can be accomplished in RGB using Luminosity blending in PS with far less damage to the image and faster too.

If we're talking Raw here, Lab and CMYK haven't entered into the picture, we are at the rendering stage. Least we forget we're handling Grayscale data. Lets render the image as closely as possible to our ultimate goals, at least globally.

As far as exposure and saturation, we're dealing with rendering scene to output referred. That's why we have sliders to control the rendering. If the image appears to need a saturation boost, do it. Its not matching the scene, it never will (and when you do view scene referred imagery output referred, it looks butt ugly). We shoot and process Raw so we can control how we wish to express how we recall the scene or want to represent the scene using the controls provided in the converter:

http://www.color.org/ICC_white_paper_20_Digital_photography_color_manag ement_basics.pdf

And

http://wwwimages.adobe.com/www.adobe.com/products/photoshop/family/prophotographer/pdfs/pscs3_renderprint.pdf

[Cem_Usakligil]

Hi Ray,

Actually, sorry to disagree, sensors are not non-linear (as mentioned at the bottom of the referenced LL article). The sensor response is almost perfectly linear to incident light, and the Raw data usually has an equally linear gamma. The Leica M8 'Raw' data is a known exception to the rule. The human eye is far from linear, it more resembles a logarithmic or a power function.

Therefore, there is little reason to expect a color/saturation shift with a uniform scaling operation (exposure slider) before other tonescaling, unless this is a side effect of Adobe's ACR color-management approach (which is initially in Raw more matrix based than LUT profile based).

Bart
I, too, agree with Bart. Sensors are linear. Their analog output is linear, meaning that the output doubles each time the light falling on the sensor doubles. Analog to Digital Conversion (ADC) is also linear. Just consider the following example (figures are fully fictional):
1: Lowest light falling on the sensor (above the SNR treshold) - sensor output: 1 mV
2: Light doubles: - sensor output 2 mV
3: Light doubles: - sensor output 4 mV
4: Light doubles: - sensor output 8 mV
5: Light doubles: - sensor output 16 mV
6: Light doubles: - sensor output 32 mV
7: Light doubles: - sensor output 64 mV
8: Light doubles: - sensor output 128 mV
9: Light doubles: - sensor output 256 mV
10: Light doubles: - sensor output 1024 mV
11: Light doubles: sensor saturated, clipping occurs - sensor output 1024 mV

In this example, if the ADC is based on a resolution of 12 bits (values between 0-4095), then the first light zone will have a resolution of 4 bits, the next one 8 bits, etc, and the 10th one 2048 bits.

So in summary, the sensors are linear (mostly). It is the intensity of light that doubles which is not interpreted linearly by our own eyes.

Cheers,

[Michael Fontana]

Doesn't it depends on the RC's internal colorspace - when you edit?

Just a example: Raw Developer (mac only) allows to tweak in LAB; the L channel can be edited independently from the A & B; therefore no color shifts.

Most of the other converters don't allow tweakings in LAB; but a linear RGB-space: so one might catch some shifts.

Is this wrong?

[Ray West]

OOPs.....

I got carried away with my own thing, which Nick's link answered. I should probably have said that the raw file output is not linear, or something like. iow, if you slide the exposure setting in ACR, it is not the same as sliding it in CS2. This has little to do with Eric's situation. wrt Canon, I wonder if dpp does it right, non-canon stuff get it wrong

Best wishes,

Ray

[Nick Rains]

As usual, Andrew has it spot on.

If you need to make an adjustment, do it. That's why we shoot RAW. If one adjustment alters something else, then correct for it - there are plenty of sliders in ACR4.2!

It's all about capturing the best data from the scene and then making it fit your vision; IOW correcting it towards a scene referred end point. ETTR ensures the maximum number of greyscale tones describe the scene and this is a Good Thing.

I think people tend to mistake this method with the classic Zone System which is not really appropriate when dealing with linear data capture.

Cheers

[Eric Hiss]

Guys,
Thanks for all your responses. 1st let me say that yes I understand that sensors are linear and that the brightest stop contains more levels. I also understand that if you push up shadows you often get ugly noise. I would have expected that with the same lighting and no clipping on either end, a file push down 1 stop would look the same as a file pushed up 1 stop. But not so.

Here's how I found this out. I went out yesterday to casually shoot some flowers yesterday. In the first shot of the day I accidentally left the camera in manual mode since I had been in the studio. When I saw the file in the display I realized it was too dark, figured out the problem and set the camera to Aperture priority mode. Retook the same shot. Okay now back at home I looked at the files, pushed up the dark one, and ended up pushing down the other using exposure control in lightroom. Okay now they both have the same white and black points but they don't look the same. The underexposed image has much more saturated color and seemingly more contrast. I thought about this and realized that a whole lot of my files were looking better exposed in the middle than exposed at the right. Hmm.....

Okay so this is with the DMR - its has a higher bit file than the canons - and I don't see a lot of problems with the shadows.

So what's the story? My guess is that when you use the exposure tool, it is in reality like curves in photoshop. If you push up, you get more saturation unless you set the blending mode to luminosity only. If you push down using curves then you get less saturation unless you also set the blending mode to luminosity only.

So my guess is that the RAW conversion tools are doing something like this with the exposure adjustment. Any thoughts?

[Don Lashier]

> My guess is that when you use the exposure tool, it is in reality like curves in photoshop. If you push up, you get more saturation unless you set the blending mode to luminosity only. If you push down using curves then you get less saturation unless you also set the blending mode to luminosity only.

Raw converter EC should not be doing this as it should be linear scaling prior to application of tone curve etc. which might affect saturation. Which RC are you using again?

- DL

[Nick Rains]

Please don't think me flippant but I suspect you are looking for technical colour problems where non exist. If you expose and process two images differently you should expect them to look different. If you like one better than the other then fine. However you need to offset this against the cost of doing so.

In film days if you underexposed and over processed you got more contrast and saturation too - at a cost of more grain.

[Eric Hiss]

I not complaining or looking for problems....just wanted to know why I was seeing what I do. If I can understand what is happening then I can either avoid it or use it to my advantage.

[Nick Rains]

OK, fair enough.

My feeling is that you should be able to achieve the same look with other controls. In theory the cost (noise, posterisation) of what you are doing will outweigh the advantages of other methods.

I can't explain 'why' you see what you do, maybe some of the more technical folk can help.

[StuartRae]

Hi Bart,

I think Nick's link sort of explains. The sensors are non-linear in their response.

Hi Ray,

Actually, sorry to disagree, sensors are not non-linear (as mentioned at the bottom of the referenced LL article).

Please blame me for this misunderstanding. In a moment of madness I mentioned to Ray that sensors were non-linear, and the poor trusting soul believed me. Of course they are not.......

Regards,

Stuart

[Colleen Vermillion]

So what's the story? My guess is that when you use the exposure tool, it is in reality like curves in photoshop. If you push up, you get more saturation unless you set the blending mode to luminosity only. If you push down using curves then you get less saturation unless you also set the blending mode to luminosity only.

So my guess is that the RAW conversion tools are doing something like this with the exposure adjustment. Any thoughts?

I think the missing piece of the puzzle is the conversion of the sensor values to RGB when shooting RAW. Cameras have a tone curve associated with them that is generally non-linear. Different RAW converters handle this differently, but the conversion into RGB almost always happens prior to the exposure adjustment. In Bibble, we have a drop down that allows you to select among different tone curves, from flat (or colorimetric) to high contrast. In ACR, the equivalent is on the "Curve" tab I think. The camera's tone curve is probably optimized for exposing correctly, so if you consistently overexpose and don't override the camera's tone curve in your processing, your mid tones are going to be pushed into the highlight "shoulder" and have their contrast lowered.

Colleen

[Bart_van_der_Wolf]

I think the missing piece of the puzzle is the conversion of the sensor values to RGB when shooting RAW. Cameras have a tone curve associated with them that is generally non-linear.

Perhaps I misread what you are saying, but the cameras themselves (their sensor arrays) operate in a strictly linear gamma fashion. This is also reflected in their Raw data after AD conversion/quanization. The Leica M8, and maybe Eric's DMR as well, apply a gamma adjustment via a LUT before saving to 'Raw'.

Different RAW converters handle this differently, but the conversion into RGB almost always happens prior to the exposure adjustment.

With "conversion into RGB" do you mean Bayer CFA demosaicing, or that plus gamma adjustment plus tonecurve adjustment? Changing 'exposure' on anything non-linear seems unnecessarily complicated and error prone.

My tests of .CR2 conversions with ACR don't exhibit shifts in saturation when 'pulling' exposure (in case of having 'over'-exposed a low dynamic range (LDR) scene to reduce noise).

Bart

[Eric Hiss]

I think the missing piece of the puzzle is the conversion of the sensor values to RGB when shooting RAW. Cameras have a tone curve associated with them that is generally non-linear. Different RAW converters handle this differently, but the conversion into RGB almost always happens prior to the exposure adjustment. In Bibble, we have a drop down that allows you to select among different tone curves, from flat (or colorimetric) to high contrast. In ACR, the equivalent is on the "Curve" tab I think. The camera's tone curve is probably optimized for exposing correctly, so if you consistently overexpose and don't override the camera's tone curve in your processing, your mid tones are going to be pushed into the highlight "shoulder" and have their contrast lowered.

Colleen

Colleen,
Thanks for you cogent response to my question. So here is one follow up question. If I set the RAW converter tone curve to linear, then push up or down my exposure then they will not affect the color saturation or contrast, but then my image will look flat. If I use ETTR to minimize shadow noise (say adding +1 stop) what should the RAW tone curve look like in post?
Thanks,
Eric

[Colleen Vermillion]

Perhaps I misread what you are saying, but the cameras themselves (their sensor arrays) operate in a strictly linear gamma fashion.
Sure, they are linearly recording sensor values. Sensor values are not RGB values... they still have to be demosaiced, converted from the camera's color space, and have the camera's tone curve applied. RAW files are not images (in the photographic sense) until those steps are taken. Exposure compensation properly happens after the file is rendered into an image, and is usually done in LAB space. To get the RAW sensor values into LAB, you have to map them. How they get mapped depends on the software that converted the RAW files into images. I suppose a dark room simile might be that the tone curve is part of the film development, and the exposure compensation is part of printing the negative (at least in B&W, I never did any color film processing)

Here's the difference between a flat tone curve:

http://home.earthlink.net/~cevermillion/images/tonecurve/FlatToneCurve.jpg

And a high contrast tone curve:

http://home.earthlink.net/~cevermillion/images/tonecurve/HighContrastToneCurve.jpg

I also put +1 exposure adjustments for each tone curve in that folder (http://home.earthlink.net/~cevermillion/images/tonecurve/). The only difference between those two files is the tone curve I selected - I didn't even do auto levels. Now, I'm not that familiar with ACR, so I can't tell you if the saturation difference really is due to the tone curve, but it is something to consider.

Colleen

[Bart_van_der_Wolf]

If I set the RAW converter tone curve to linear, then push up or down my exposure then they will not affect the color saturation or contrast, but then my image will look flat.

That's because of the difference between photometrically correct Raw conversion, and what human perception needs/expects. One should add a tonemapping operation for perceptally pleasing images. A variation of the "large radius USM with low amount" adaptive method already helps to get a more pleasing contrast.

The following link http://www.xs4all.nl/~tindeman/raw/curve_tools.html (http://www.xs4all.nl/~tindeman/raw/curve_tools.html) is also relevant to this thread.

Bart

[Colleen Vermillion]

If I use ETTR to minimize shadow noise (say adding +1 stop) what should the RAW tone curve look like in post?

The tone curve is going to depend on your personal taste, the subject matter, and possibly the color management if you're shooting content that can go out of gamut (like stage lighting). I think that a linear curve can always be compensated for after the fact, but it might be unnecessary work if you are going to be consistently shooting the same way. I'm a "whatever gives you the results you like" sort of person... experiment with different curves with a set of your typical scenes and see how it feels for you.

-Colleen

[Andrew Rodney]

With respect to curves (in an Adobe Raw converter or Photoshop) and saturation, you might what to read this:

http://www.luminous-landscape.com/essays/curves.shtml

[John Sheehy]

Okay I have always heard that with digital that expose to the right then correct in RAW conversion is the proper way to manage digital capture, however when this is done the resulting images may have nice shadow detail but are less vibrant. If you expose to the left (but not to the point of clipping the shadows) then push up the image exposure in RAW then you get a much more vibrant image.

What's happening - shouldn't the two paths result in the same image? I'm just wondering why they are not. Anyone know what's going on? I've seen this now with both Flexcolor and lightroom RAW conversion tools. Btw - I am working with Leica DMR files on images that had plenty of DR to spare on both ends. Could have exposed +- 2 stops either way and still captured all the data without clipping.

Sounds like you're leaving it to the right in PP. I set my DSLRs to minimum contrast, low saturation, and expose to the right, and the resulting JPEGs or default conversion will need a gamma adjustment of about 0.8 to return to normal, if they are a bit pale. Any compression of tonal range lowers saturation, and that's what usually happens when a conversion rolls in a lot of highlights instead of hard clipping 2.5 stops above middle grey.

[Eric Hiss]

Sounds like you're leaving it to the right in PP. I set my DSLRs to minimum contrast, low saturation, and expose to the right, and the resulting JPEGs or default conversion will need a gamma adjustment of about 0.8 to return to normal, if they are a bit pale. Any compression of tonal range lowers saturation, and that's what usually happens when a conversion rolls in a lot of highlights instead of hard clipping 2.5 stops above middle grey.

Nope, I think Collen had it right - I was not setting the RAW conversion to linear for the ETTR shots. Still it is difficult to get the same image as normal exposure with ETTR. btw-I also shoot MF digital and there is a long thread on this at the Luminous Landscape forum.

[Gabor Schreiner]

Eric,

can you post/send me such a raw file, on which this de-saturation problem can be demonstrated?

[Eric Hiss]

Eric,

can you post/send me such a raw file, on which this de-saturation problem can be demonstrated?

No but when you finally get your own camera (your 1st post here states you don't have one and this request is your 2nd and last post) you can easily test this out yourself.

[Nicolas Claris]

No but when you finally get your own camera (your 1st post here states you don't have one and this request is your 2nd and last post) you can easily test this out yourself.

Eric
in the post you're reffering, Gabor is stating that he does not have the camera involved in the thread, not that he doesn't have any…
In the same post he states that he's a pixel peeper, maybe he has a good knowledge that could help you to understand the issue you're talking about in this thead… who knows?

this request is your 2nd and last post
and then? is that a reason for not answering the request?

I do understand and completely agree that there could be a lot of good reasons not to attach or send a raw file… (client shot, commercial, special technics used you want to keep confidential, etc.) but the reasons you give here are not good enough for me and not very fair toward Gabor…

[Eric Hiss]

Nicolas,
Your post is coming through as double.

I stand by my comments. The requester appears to be only asking for self satisfying reasons, and didn't even say please.

[Nicolas Claris]

Nicolas,
Your post is coming through as double.

Thanks, deleted one…

[Gabor Schreiner]

I stand by my comments. The requester appears to be only asking for self satisfying reasons, and didn't even say please.

Well, the initial post of this thread contained

What's happening - shouldn't the two paths result in the same image? I'm just wondering why they are not. Anyone know what's going on?

What kind of help are you expecting? Without being able to reproduce the effect, one can only speculate. Reproducing the effect requires having the raw images and knowing the exact adjustments in ACR. Furthermore, a low-level analysis of the raw data is quite difficult without actually having it.

Anway, good luck.

[Eric Hiss]

Gabor,
If you read the thread you will see that I already got a satisfactory answer from Colleen. Don't need any further info. There is no more speculation, however I encourage you to experiment to your hearts content with your own images.
Eric

[John Sheehy]

Okay I have always heard that with digital that expose to the right then correct in RAW conversion is the proper way to manage digital capture, however when this is done the resulting images may have nice shadow detail but are less vibrant. If you expose to the left (but not to the point of clipping the shadows) then push up the image exposure in RAW then you get a much more vibrant image.

What's happening - shouldn't the two paths result in the same image? I'm just wondering why they are not. Anyone know what's going on? I've seen this now with both Flexcolor and lightroom RAW conversion tools. Btw - I am working with Leica DMR files on images that had plenty of DR to spare on both ends. Could have exposed +- 2 stops either way and still captured all the data without clipping.

In the RAW data, the only difference is the amount of noise (and technically, the number of levels per stop, but that advantage is completely obscured by noise). If the exposure slider in your converter isn't exactly reversing the exposure, then that exposure slider is not a true linear scaling slider, and in my experience, most aren't.

[Simon Tindemans]

What's happening - shouldn't the two paths result in the same image? I'm just wondering why they are not. Anyone know what's going on? I've seen this now with both Flexcolor and lightroom RAW conversion tools.

In the case of Lightroom, the answer could lie in the 'blacks' slider. It defines a constant amount that is subtracted from all channels before the exposure slider is applied.

At its default setting of 5, it could easily explain the phenomenon you are seeing. If you overexpose and pull back in processing, you effectively get a lower value, leading to less saturated colors. If you underexpose and push in processing you effectively get a higher value, leading to more saturated colors.

Try setting the blacks slider to zero and see whether the problem disappears.

As an aside, to me this indicates that the blacks slider is serving two roles, and it seems misplaced for one of them. The first role is flare reduction. That should indeed be done in the initial stages of the pipeline, but the default value should probably be zero. The second is a black level control for the output-referred image with a default value that corresponds to the current default of 5. The latter should really be combined with the tone curve stage, IMO.

Simon

PS - I just noticed that this thread has been sleeping for a couple of months. I figure the OP may not be on this anymore, but as my answer would still be valid, I'll leave it here.