exposure vs color/tonality with digital cameras

[button]

If a digital capture is acquired at a certain EV, will the identical image (control for lighting, framing, etc) have the same color range/tonality if exposed at a different EV and then corrected (pushed or pulled)?  Ignore noise, and assume no blown highlights/shadows.

John

[Tony Beach]

If you are shooting in RAW and using a good RAW converter, then yes; for JPEG probably not.

[bjanes]

If a digital capture is acquired at a certain EV, will the identical image (control for lighting, framing, etc) have the same color range/tonality if exposed at a different EV and then corrected (pushed or pulled)?  Ignore noise, and assume no blown highlights/shadows.

John

For each doubling of the ISO, you lose up to 1 stop of dynamic range, as shown in Table Ib in Roger Clark's analysis. If you use the engineering definition of dynamic range, the DR is the number of electrons collected divided by the read noise. If you double the ISO, you collect half the number of electrons. For the Canon 1DMII near base ISO, there is initially not much loss in DR as the fewer electrons collected is offset be a decrease in the read noise. For this camera halving of DR with doubling of ISO is observed at or above ISO 800.

If you expose for the highlights at high ISO, there will have to be clipping in the shadows with a high dynamic range scene; this is implicit in the definition of DR.

At very high ISO color rendering may also suffer.

Bill

[Guillermo Luijk]

If you are shooting in RAW and using a good RAW converter, then yes; for JPEG probably not.

A RAW developer has to be certainly very badly designed to change Hue/Saturation when applying an exposure correction, since this is a trivial math operation: just multiply the 3 RGB channels by the same scaling factor.
Even once the image has been converted to a colour profile and gamma has been applied, correcting exposure with no hue/sat change through scaling remains true as long as the profile is linear (usually) and the gamma is a true gamma (this excludes sRGB).

Curves such as these can be used in Photoshop to perform exposure correction even over JPEG files:


(the +1EV/-1EV values would apply if gamma=1.0)

The only difference after exposure correction (leaving aside the possible blowing of the highlights) will be the lower presence of noise when RAW exposure was higher:

[Dale_Cotton2]

The aspect that would concern me comes from the gospel according to ETTR. If you place one exposure a stop lower than another, you've halved the number of potential hue gradation bits for each channel of each stop/zone in the image.

[Guillermo Luijk]

The aspect that would concern me comes from the gospel according to ETTR. If you place one exposure a stop lower than another, you've halved the number of potential hue gradation bits for each channel of each stop/zone in the image.

Every extra EV of RAW exposure means having twice as many levels. However, this effect is almost negligible since in the deep shadows, where this could be a real advantage, noise appears before we actually can enjoy the advantage of those extra levels. Even 12-bit sensors (with 16-bit on interpolated channels) have a soft enough tone gradation that ETTR does not mean a benefit regarding tonal richness.

Improved SNR is the only visible result as today's sensors are still noise-limited. That is why the introduction of the last 14-bit cameras didn't mean any improvement just for the increased bitdepth, but for other reasons.

[Dale_Cotton2]

GLuijk: Thanks for the quick response. I follow your first point just fine, re shadow vs. noise. But then you lose me when you write:

Even 12-bit sensors (with 16-bit on interpolated channels) have a soft enough tone gradation that ETTR does not mean a benefit regarding tonal richness.

What I don't understand is what you mean by "soft enough tone gradation". If a certain stop has 4096 gradations per channel and the stop below it has 2048 gradations, are you saying that the 4096 stop will not produce more visible hues than the 2048 stop?
*
As an aside, I greatly enjoy and appreciate it whenever you, Emil, and other technical experts take the time to post. As a programmer by day and an artist by night, it's that darned electronic circuitry in the middle that keeps trying to trip me up.

[button]

Thanks, everyone- this is helpful.  A potential application, if I'm understanding this discussion: shooting pano/mosaics in Av mode at fixed ISO, as suggested in this thread:

http://luminous-landscape.com/forum/index....showtopic=28921

Any comments?

[vandevanterSH]

(snip)
"A common maxim in digital photography is that image quality is maximized by "exposing to the right" (ETTR) -- that is, raising the exposure as much as possible without clipping highlights. It is often stated that in doing so, one makes the best use of the "number of available levels" in the raw data. This explication for instance can be found in a much-quoted tutorial on Luminous-Landscape.com. The thinking is that, because raw is a linear capture medium, each higher stop in exposure accesses the next higher bit in the digital data, and twice as many raw levels are used in encoding the raw capture. For instance, in a 12-bit file, the highest stop of exposure has 2048 levels, the next highest stop 1024 levels, the one below that 512 levels, and so on. Naively it would seem obvious that the highest quality image data would arise from concentrating the image histogram in the higher exposure zones, where the abundance of levels allows finer tonal transitions.

However, the issue is not the number of raw levels in any given segment of the raw data (as measured e.g. in stops down from raw saturation point). Rather, the point is that by exposing to the right, one achieves a higher signal to noise ratio in the raw data. The number of available raw levels has little to do with the proper reason to expose right, since as we have seen the noise rises with signal and in fact the many raw levels available in higher exposure zones are largely wasted in digitizing photon shot noise (there will be more to say about this in a moment, when we consider NEF compression). "
(snip)
http://theory.uchicago.edu/~ejm/pix/20d/te...oise/index.html
************
It's over my head but it is an interesting reference.

Steve

[Guillermo Luijk]

If a certain stop has 4096 gradations per channel and the stop below it has 2048 gradations, are you saying that the 4096 stop will not produce more visible hues than the 2048 stop?

Yes it will, but you will not see them since your eyes cannot differentiate so many levels, so you will not enjoy that advantage. Think that captured samples are 12 (or 14) bit, but:
First: they are affected by noise, which in the shadows is greater than one level step (see Emil's Fig. 8 in: Noise, Dynamic Range and Bit Depth in Digital SLRs)
Second: the other two channels on each pixel are interpolated in 16 bit, so their tonal richness will provide soft gradations in hue (even if that does not mean they have 16 bit precision of course, since they are just interpolations from 12-14 bit values).

So the really important thing of ETTR is noise reduction, not the 'more information' claim.
For example the Canon 5D's RAW files, a fantastic camera and sensor, start at level 128-256, and saturate at 3692. That means this camera has just around 3500 levels, quite less than 4096 and much less than any 14 bit camera. Did you hear of some 5D user complaining of banding, posterization or something due to the lack of levels of his RAW files?
Another example: the Leica M8, a quite expensive top camera, has RAW files of just 8 bits, i.e. 256 levels. They are simply arranged in a very clever non-linear way to compensate the lack of levels in the shadows by substracting some unnecesary precision from the highlights. But the RAW files themselves have the same gradation as a JPEG file (8 bits).

However, the issue is not the number of raw levels in any given segment of the raw data (as measured e.g. in stops down from raw saturation point). Rather, the point is that by exposing to the right, one achieves a higher signal to noise ratio in the raw data.

Exactly.

BR

[Panopeeper]

The claim, that low number of tonal gradations do not cause banding/posterisation is unfounded speculation. The fact, that some images do not exhibit posterisation, even that the majority of images do not exhibit posterisation is no proof. The proof is in the pudding: if some images do exhibit banding/posterisation due to the lack of fine gradation, that's enough.

As I have seen such images, I can say with certainty, that yes, the lack of gradations can pose a problem. I analyzed a Nikon D300 image a short while ago; the photographer has complained about banding in the sky at ISO 200. The data was recorded in 14bit mode, but with lossy compression. As the setting was repeatable, the photographer has reshot the scene, with both lossy and lossless compression; this experiment has proven clearly, that the  lossiness was the source of banding. (Note, that a banding appeared even with the lossless data, but to a much smaller degree.)

Even Nikon are cautios with the lossiness; they are stating that it "does not cause any problem in the majority of cases" or something like that.

[Panopeeper]

A potential application, if I'm understanding this discussion: shooting pano/mosaics in Av mode at fixed ISO, as suggested in this thread:

http://luminous-landscape.com/forum/index....showtopic=28921

I would be very cautios about that. I have shot many panos with variable exposure, in order to increase the dynamic range of the overall result. Sometimes it works well, but in some cases it requires extraordinary effort to match the frames. (I never use the correction feature of the stitcher.)

In difficult cases one may have to create a graduated transition of some (or each) frame to match it to the adjacent frames. It is doable, but one has to be prepared for much work. There are some panos of mine, which required dozens of hours (each) to finish them.

[Panopeeper]

If a digital capture is acquired at a certain EV, will the identical image (control for lighting, framing, etc) have the same color range/tonality if exposed at a different EV and then corrected (pushed or pulled)?  Ignore noise, and assume no blown highlights/shadows.

It should not matter, but it may. My former favourite raw converter, RSE, did not make a really linear adjustment and that caused color mismatch between the frames of my panos, when shot with variable exposure; I had to use DPP in such cases.

[Guillermo Luijk]

Gabor, could you please show examples of banding in the skies due to not fine enough gradation? Banding in skies is a subject in which I am very interested, and surprised too, after seeing correctly exposed (I mean highly enough exposed) RAW files that still produced easily banding in the sky. Even I know of some professional arquitecture photographers that usually add noise in their skies to get a soft gradation.

I could never find an explanation to this (a RAW with a good exposure showing banding in the sky), and in fact these skies improved when exposure was reduced thanks to the dithering effect of the extra noise. This made me think these skies were very sensible to banding, even if there should be a fine enough gradation in the B channel.

This is an old test I did: the scene was like this:




Taking the 100% crops on the left side (where the blue gradation was softer), matching exposures and applying curves to peep gradation:

Correctly exposed RAW file:


ETTR RAW file +2V with respect to the former:



The added noise in the first case seems to help make the gradation most robust against posterization (I don't mean that noise is desirable with this). I am really puzzled with some skies where the more you expose seems the worse results you get. If we eliminated noise, both exposures ended in the same main bands even if one was 2 stops higher in exposure (and hence captured 4 times more differentiated RAW levels):




It seems the ETTR didn't mean a softer gradation, just less noise!.

BR

[Panopeeper]

Gabor, could you please show examples of banding in the skies due to not fine enough gradation?

Lossy 14bit NEF
url="http://www.panopeeper.com/Download/NikonD300_Banding_14bit_lossless.NEF"]Lossless 14bit NEF[/url]

Lossy JPEG crop
url="http://www.panopeeper.com/Demo/NikonD300_Banding_14bit_lossless.jpg"]Lossless JPEG crop[/url]

I too am interested to see your sample raws.

[Guillermo Luijk]

I too am interested to see your sample raws.

It's an old test almost 2 years ago, I don't keep the files sorry. I will try to repeat it when I have more time.
Anyway this is an issue commonly reported in arquitecture forums, most of those photographers face this problem. Didn't you ever see that in any of your landscapes?

[skid00skid00]

It seems the ETTR didn't mean a softer gradation, just less noise!.

The banding showing in the crops you posted is the RESULT of appying a strong curve to a gradient.  It will happen to any smooth gradient, irregardless of source.  It is not the result of banding in the source image.

[Guillermo Luijk]

The banding showing in the crops you posted is the RESULT of appying a strong curve to a gradient.  It will happen to any smooth gradient, irregardless of source.  It is not the result of banding in the source image.

That is clear, the question is why the 2 extra stops didn't mean an improvement in the basic bands of this banding. I would expect 4 times more bands in the second shot meaning softer transitions.

[button]

I would be very cautios about that. I have shot many panos with variable exposure, in order to increase the dynamic range of the overall result. Sometimes it works well, but in some cases it requires extraordinary effort to match the frames. (I never use the correction feature of the stitcher.)

In difficult cases one may have to create a graduated transition of some (or each) frame to match it to the adjacent frames. It is doable, but one has to be prepared for much work. There are some panos of mine, which required dozens of hours (each) to finish them.

The reason I started this thread has mostly to do with my wanting to expedite my capture technique for stitching.  The last time I used Av mode for a mosaic, I spent quite a bit of time in photoshop adjusting each frame so that the skies matched- I had to use custom curve adjustment layers, each of which required points in the RGB curves.  Now, I don't know about the rest of you, but I think that's way too much time to spend.  

I had hoped that by simply adjusting the exposure in ACR to compensate for the differences in capture exposure that I could even out the sky.  Unfortunately, that didn't work, which made me wonder about the possible effect of exposure on color/tonality (yes, I took the shots in rapid succession; yes, I synchronized the WB on the RAW files; and no, I didn't use a polarizer).  Can anyone think of other variables to consider with regard to this issue?

[Panopeeper]

The last time I used Av mode for a I had hoped that by simply adjusting the exposure in ACR to compensate for the differences in capture exposure that I could even out the sky.  Unfortunately, that didn't work, which made me wonder about the possible effect of exposure on color/tonality (yes, I took the shots in rapid succession; yes, I synchronized the WB on the RAW files; and no, I didn't use a polarizer).  Can anyone think of other variables to consider with regard to this issue?

This is not a simple issue.

First think of the reason you shoot with variable exposure. Typical scenery: as you turn, the camera is turning into or away from the sun; the ideal exposure changes gradually. A given exposure will blow one frame, while another one gets underexposed. So, you shoot it with variable exposure, then in PP you want to equalize the frames. This means, that you pick one, perhaps the one with a medium exposure, and adjust the neighbouring frames to this one. Thus in effect you reverse the result of the variable exposure. Now go to the next frame and do this; the adjustment has to be much more. So, you end up with a set of frames, which look exactly (in ideal case) like you would have shot them with fixed exposure.

I that, what you wanted? Most probably not. Though this can help on noise issues, that's not all. You want to lighten those frames, which would be too dark with the fixed exposure and/or darken those, which would have been overexposed.

In effect, you want to adjust the overall intensity over all frames continuously. This requires the adjustment of the frames pre- or post stitching (but before blending), gradually. You need a graduated mask for the adjustment, "bridging" over the frame, "connecting" the two adjacent frames. This is, what I ment with "extraordinary effort".

Another, cheaper way is to adjust the frames for example in ACR not with "exposure" adjustment but with fill light, recovery and/or brightness. You can match the sky (more or less) this way, and ignore for example a landscape, for it is not as sensitive to differences as the sky. (Note: a water surface is almost as sensitive as the sky). However, now the colors of the frames will go apart; only the grey remains grey.