Luminous Landscape Forum
Equipment & Techniques => Digital Cameras & Shooting Techniques => Topic started by: bwana on April 06, 2012, 11:20:02 pm
-
As I understand it, most histograms displayed on the back of a camera LCD are based on the jpg derivation of the raw image. Since the jpg is a compressed version of the raw, then to really go to the right edge of the raw histogram, you have to clip a little in the jpeg?
In addition, cameras are coming with more and more controls to change the processing in camera. The latest olympus micro 4/3 cameras even allow application of tone curves. Would these manipulations also not skew the histogram and therefore distort where ETTR should be?
Then again, this 'inability to know where the right edge of the raw histogram really is might actually be a bonus. It is the 'insurance' that allows us to recover highlights. So although we do not expose right up to the edge, we get out of the dark noise areas. So how far to the right do we REALLY need to go?
Maybe we dont need to go as far to the right any more. I guess it really depends when the S/N ratio comes out of the darkness. Really you just have to get the histogram out of the the low S/N region on the left side. I might still have a lot of headroom on the right side (as sensors come out with greater dynamic range).
Is there a program that will calculate S/N of a camera sensor from an exposure of a ramp of dark tones? But then you might say, what good is a number like S/N anyway. What does it really mean?
For S/N to have any meaning, we have to know what the human eye can perceive. Keeping S/N above that threshold in the raw capture would be the goal.
-
What does it really mean?
You shouldn't rely on the camera histogram for anything more than a ruff guide.
And bracketing still works.
-
There is no absolute answer to this question, no equation that can just simply be applied and hey presto the perfect ETTR exposure.
How to expose to the principles of ETTR are somewhat dependent on the performance of the sensor as already suggested, nonetheless individual scenes will have a large impact on what determines the actual exposure.
Specular highlights are a case in point. Does one want to include the specular highlights in determining how to expose or is it OK to let them blow.
How important are other highlights in the composition or is OK to let them blow.
No equation can determine this only the goal and vision of the photographer can determine how to expose for the specific scene in question.
No doubt that when one uses a particular camera for a time one does get a feel for how far to push the exposure but I am absolutely with K.C. when he suggests bracketing exposures around an apparent ideal exposure. Only round-tripping between the field and ones computer can one start to get a real idea of ETTR. There is also a learning curve associated with post-processing ETTR images. One may get the impression that ETTR does not work if the post-processing is not good.
Additionally, even if using the same hardware, my judgement of an ideal ETTR exposure may not square with anyone elses. Although the principles of ETTR are based on sound principles of physics ones own artistic vision and tolerances will actually determine how one exposes for ETTR.
Regards
Tony Jay
-
After some practice you can judge how many "blinkies" on your rear screen jpeg correlates to RAW clipping.
Just shoot a lot of ETTR's and compare your raw histograms that are ETTR'd just right to how severe the "blinkies" are prior to clipping. About the best you can do without a RAW histogram displayed. As mentioned several exposures increases your odds. Just be careful on those blue skies the blue channel might clip before the others and your "blinkies" might not be an accurate representation.
Marc
-
In additional to knowing how far you can or should go with your exposure, you should also set the JPG processing parameters to obtain the most neutral processing possible. I think this means, at least, minimum contrast and sharpening.
Search the forum for ETTR. There have have been many discussions of how best the set those parameters.
Alan
-
Take a shot with the histogram right up to the edge then load it into your raw converter with usual starting settings and see how exposure you need to increase before it clips. Personally, I just use this as insurance
-
The histogram generated by the camera is a representation of a processed JPEG not the RAW file.
In order for the camera histogram to more closely represent the RAW file change settings such as contrast to zero (in most cameras changing the picture style to neutral should do the trick).
Nonetheless the camera histogram does not exactly match what the RAW file has captured.
Camera manufacturers have been lobbied (not least by Michael Reichman himself) to change to, or at least allow the option of viewing, a histogram representing a RAW file. To date, no dice.
So only once a RAW image has been loaded into a RAW converter can the dynamic range of the image really be appreciated.
Hope I am not selling snow to eskimos
Tony Jay
-
'How do you really know when you are ETTR?'
Unfortunately you can only know this once at home, and analysing your RAW files with an appropiate piece of software (Rawnalyze, RAW Digger, DCRAW,...).
At shooting time in the camera or at home using conventional RAW developers, today it's impossible to make sure if you ETTR'ed or not.
Regards
-
Hi,
I agree with Guillermo, but in my view it's more about a mindset than achieving the absolutely correct exposure. The mindset is that we really want to maximize exposure. My experience with my equipment (Sony) is that blinking highlights and histogram are quite OK. Each time I went over blinking highlights and clipped the histogram I got burned out highlights.
Best regards
Erik
'How do you really know when you are ETTR?'
Unfortunately you can only know this once at home, and analysing your RAW files with an appropiate piece of software (Rawnalyze, RAW Digger, DCRAW,...).
At shooting time in the camera or at home using conventional RAW developers, today it's impossible to make sure if you ETTR'ed or not.
Regards
-
thank you all for taking the time to answer the first part of this question. but the flip side of the coin still nags me. to put it simply, how far to the right do i have to go to get out of the noise regime of my sensor?
i know this varies depending on the sensor obviously- but i dont want to buy DXOanalyzer to check my camera.and even if i did, i wouldnt know how to translate those numbers into a useful guide. here for example is an evaluation of the canon 1D mark2
http://www.clarkvision.com/articles/evaluation-1d2/
S/N at iso 50 is 283, at iso 400 it's 115
how much do i have to adjust the EV to get clean data out of an image of a dirty,dark closet? If on my histogram on the back of the camera, 255 is white and 0 is black, what should be my lower limit where my darkest black should be captured?The goal is so that when i adjust the raw in PP and make that black 0, there is no noise.
now i know some of you will say that i might be clipping highlights, but i am not focussed on that at the moment. assume the image is low key or that multiple exposures with hdr technique will be used to capture the upper end.
and i did google extensively before asking this.
http://lmgtfy.com/?q=measure+camera+sensor+noise
but i am searching for meaning, not just information
(i think i should add that to my signature)
-
Your camera sensor is a tri-band photon counter. This means it averages the number of photons coming from 3 areas of the spectrum, one around the red wavelengths, one around the green wavelengths and one around the blue wavelengts.
Perfect ETTR would mean no information gets clipped in any of the three channels, and this depends on the scene. Most scenes will have a dominant G channel, but some others will have R or even B dominant channels. So unless your camera provides you with RAW histograms (which unfortunately is a very silly thing to achieve, but camera makes are not interested in), you cannot figure out a simple rule of thumb appicable to all cases. This thread (http://www.luminous-landscape.com/forum/index.php?topic=46071.0) shows why.
The good part is that cameras get more and more DR with improved sensors, so perfect ETTR is less and less necessary with time. If you make sure your JPEG has no clipped areas, or minimal clipped areas, you can be 99% sure your RAW file is intact.
-
If the camera histogram is generated by a processed jpeg, how does that work on a digital back that only produces RAW files? Is there a difference? It is discouraging to find right clippings in post processing software when one sees none in the camera in the field.
-
I have to confess that I do not know the mode of generation of MFDB 'in camera' histograms.
As far as I am aware with regard to 35mm equivalent DSLR's my statement was accurate.
I do feel that camera manufacturers need to give photographers at least the option of viewing a histogram representative of the RAW capture.
Regards
Tony Jay
-
http://www.brisk.org.uk/photog/histo5.html
This link explains some of the limitations of in camera histograms and their analysis.
It is a bit dated so newer information may be available.
Regards
Tony Jay
-
I dont feel I am getting the answer I am looking for , probably because I am not communicating well. Consider driving down a road to avoid an obstacle. What I am hearing from most of you is to watch out for the curb on the right side of the road and go as close to it as possible. (Apologies to those of you in the UK, just pretend you are driving in Europe) I am more interested in focussing on the obstacle to avoid it. What this translates to is a vertical guideline on the histogram chart that marks where the left side of the histogram *should* end so there is not perceptible noise. I know that noise is constant across the entire exposure range but it is just noticeable more in dark scenes. Only when the brightness of the signal exceeds the brightness of the noise does the noise become visually insignificant. This is a specific point on the horizontal axis of the histogram and is related to the sensor. Can this point be measured? Tnx.
-
What I am hearing from most of you is to watch out for the curb on the right side of the road and go as close to it as possible.
Yes but this isn’t really much different from exposing transparency film. There may be highlights in a scene you wish to record, you don’t want to blow them out. You want to get as close to that as possible with ETTR. The problem is, we don’t have a good on-screen histogram of the raw data. But then we didn’t have histograms for film! If you know how your film+processing responds along with your metering techniques, you can do the job as you desire and that is much the same with digital. ETTR is what it is because the feedback on the camera isn’t accurate for the raw capture (so the recommendations are to expose to the right of this incorrect indicator).
My suggestion is to get to better know your camera system and ignore the stuff on the back of the camera in terms of exposure for raw. If in doubt, it is better to have a tad more noise than blow out highlights you wish to record. If you do that, nothing will bring that back. If possible, bracket. Find the limitations of what the correct exposure settings based on a meter provides onto the raw data along with the raw converter you’ll use to ‘normalize’ the rendering.
-
Only when the brightness of the signal exceeds the brightness of the noise does the noise become visually insignificant. This is a specific point on the horizontal axis of the histogram and is related to the sensor. Can this point be measured? Tnx.
Sorry, but no, it's not a specific point. It's hugely dependent on your perception (different people have entirely different opinions of what "visually insignificant" is), and it's also hugely dependent on post processing. E.g., with default settings the same image post processed with LR3 will look less noisy than if it were processed with LR2, just because LR3's noise processing is much better. LR4's extensive "content aware" processing adds a whole additional layer of complexity.
As suggested by Andrew, you'll do better working on understanding how your camera and workflow behaves end-to-end under various circumstances than trying to make decisions based on a theoretical ETTR threshold.
Sandy
-
Sorry, but no, it's not a specific point...
Exactly what we all have been trying to communicate the last few days!
Regards
Tony Jay
-
I dont feel I am getting the answer I am looking for , probably because I am not communicating well. Consider driving down a road to avoid an obstacle. What I am hearing from most of you is to watch out for the curb on the right side of the road and go as close to it as possible. (Apologies to those of you in the UK, just pretend you are driving in Europe) I am more interested in focussing on the obstacle to avoid it. What this translates to is a vertical guideline on the histogram chart that marks where the left side of the histogram *should* end so there is not perceptible noise. I know that noise is constant across the entire exposure range but it is just noticeable more in dark scenes. Only when the brightness of the signal exceeds the brightness of the noise does the noise become visually insignificant. This is a specific point on the horizontal axis of the histogram and is related to the sensor. Can this point be measured? Tnx.
I agree with the Digitaldog and Sandy that you are approaching the problem in the wrong way, but it is easy to use the DXO data to see how many stops a given noise floor is from saturation (100% ETTR). DXO has a full SNR plot showing the SNR in dB for a given percent of saturation. The example here is for the new Nikon D800. The Y values of 18, 24 and 30 dB correspond to SNRs of 8:1, 16:1 and 32:1 respectively. For a SNR of 18 dB, the sensor saturation is 0.16%, which is 9.27 stops from saturation (stops = log base 2 (100%/0.16%). For 24 and 30 dB, the corresponding number of stops are 7.52 and 5.52 stops respectively.
What constitutes an acceptable SNR is a personal and subjective matter. See Fig 13 of Emil Martinec's post (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p2.html#SNR-DR) for an illustration.
Unfortunately, the camera histogram does not have a log base 2 x-axis (f/stop increments) and the right end of the histogram is usually short of 100% saturation. In practice, it is best to place the highlights just short of clipping and the shadow SNR will be optimized. Because of problems with the camera histogram, one could use a spot meter to place the highlights and then measure the shadows downward from this point.
Regards,
Bill
-
Sports photography rarely allows time for precise spot metering. I still use an incident meter at times, as I did with slide film. I am not suggesting my simplistic metering/RAW exposure method is bullet proof but it gives me a good starting point and has served me well since moving from transparency to digital, some 11 years ago.
With any new camera I record a Colorchecker under daylight conditions for Lightroom camera calibration purposes, taking an incident reading as my starting point. I then include a white cotton T shirt, bracketing the exposures again by 1/3rd of a stop. Many sports involve the use of whites.
I make a note of the shutter speed and aperture that caused the first highlight warning to show on the cameras LCD. In Lightroom I flick through the images to locate the exposure which is as close to 255 as possible for my 'normal' exposure - on average this results in 1/10th of a stop shy of 255.
I note the shutter speed and aperture for this 'normal' exposure, checking it against the original Incident reading - I leave the incident meter on the original exposure reading and recalibrate (internally) to meet the same shutter speed and aperture for the 'normal' Lightroom result.
I also experiment to see if I can recover over exposed highlights (daylight conditions) I normally don't go above 2/3rds for safety - there is often one, or two channels, intact at only +2/3rds.
Using a light meter has trained me to remember the exposure off by heart in full sunlight for subjects that contain whites - film and digital. With Fuji Velvia rated at 40 ISO for my E6 processing lab I could not better 1/640 at f/4 - (600mm f/4 lens related exposure recall). With the Nikon D3S and D4 In full light, subjects with whites, if I want to stay just shy of clipping, F/4 at 1/2000 sec (100 ISO) does the trick.
I can turn down various jpeg camera settings but tend not to bother - Uni White balance is not for me. If competitors are not playing ball and wearing all black clothing (some golfers occasionally do) I have to change tactics to ETTR satisfactorily.
Important texture in whites, also caucasian skin tones from good ETTR (not clipped) stay too thin and require pulling down. Lightromm 4 is so powerful, it's the first RAW converter I have used (tried many over the years, C1 Pro, RPP, Raw Therapee, etc) which has enabled me to reduce lighter tones using a combination of the powerful whites and highlights sliders to mimic the rolled off highlights which film provides.
-
thank you for your thoughtful and considerate replies. and SO FAST. i especially appreciate your genuine sharing of your process. yes, getting to know the camera is all about iteration> capture->development->print. These articles
http://www.dxomark.com/index.php/Publications/DxOMark-Insights/Noise-characterization
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p2.html#SNR-DR
(specifically figure 12)
is what led to my question. (of course bjanes has even posted a more recent graph). I thought that the curve in the graph was the boundary that the histogram should not cross. But the horizontal axis (abcissa) is not f stops but signal output. And of course the abcissa is a log base 10, not base 2. So there is no obvious relationship between that graph and a histogram. I neglected to read the last sentence of the second article which states:
"Locate the raw value where the S/N ratio meets your threshold for acceptability, then count the number of stops to the saturation raw level; this then is the useful dynamic range. For instance, suppose the minimum acceptable S/N ratio is 4=22; on the 1D3, this S/N ratio is achieved for ISO 200 at roughly S=24.4, 9.5 stops below the saturation value of 13.9 (the max raw value is a bit less than 214-1=16383 due to the bias offset of 1024)."
So if I knew where ~9 stops to the left of the right most clipping value is, then I was thinking that would be my lower threshold (if I accepted a minimum S/N=4). (That might make a good marketing tool for the gullible- imagine Canon advertising 'noiseguard technology' simply by putting a dashed vertical line there!) But as many of you have said, there are other factors-for example, noise reduction algorithms which can selectively be applied (now in lightroom 4) to only the dark areas will spare the detail we want to preserve in the brighter areas. And there are only so many miliseconds you have before capturing 'the moment' - my thoughts should first go to preserving highlight detail than avoiding dark noise.
-
As long as the only information your camera provides is a JPEG histogram, which is affected by white balance, exposure corrections, tonal curves, colour profile conversions, saturation and sharpness enhancements,... you will not be able to translate any of all those considerations made about stops and SNR in DxO data into your real life worfkflow.
-
And there are only so many miliseconds you have before capturing 'the moment' - my thoughts should first go to preserving highlight detail than avoiding dark noise.
Whilst charts and graphs provide valuable insights, you have hit the nail on the head re preserving highlight detail - that's what a properly calibrated incident meter (colorchecker) to your choice of RAW converter can offer. Little need for 1 degree spot metering and calculations which are not always an option anyway.
Another method for fast, accurate metering, is to use the cameras metering system (manual exposures) on a nearby surface that doesn't require the narrow angle of a spot meter, provided the surface is under the same lighting as the subject. This method is useful when there is little time to pick up an incident meter, or waving one around in the air might disturb sensitive subjects - golf.
Method
Take an incident reading early on (and occasionally throughout the day when locations/lighting direction and intensity changes), apply the shutter speed and aperture indicated by the incident meter to the camera, noting where the cameras metering diode indicator settles on its exposure scale - from now on use the cameras metering system to mimic the results from the incident light meter.
For a large surface such as grass it doesn't matter which camera metering pattern is used. On grass the cameras meter will often differ from the incident reading - invariably indicating 1/3 - 2/3 rds of a stop under exposed compared to the incident reading - often depends on which way the grass has been mown!
If the light suddenly changes and there is no opportunity to use the light meter, use the cameras meter, adjusting the shutter speed or aperture so the cameras metering diode is returned to the same spot on its exposure scale as initially observed from transferring the incident metering exposure to camera - it could be over or underexposure depending on the surface measured in relation to the initial incident reading used for reference purposes.
In the UK light levels often change very quickly, this system has enabled me to quickly obtain the correct exposure (ETTR) from full sunlight to heavily diffused sunlight seconds prior to sporting moments. It is obviously useless for some sporting scenarios - e.g. rugby player crossing over from sunlight into full shade from sharp shadows cast by the spectators stand.
-
This question will probably never be settled entirely for the very good reason that the answer varies so much with the type of shooting one does.
Knowing the principles governing ETTR is exceptionally useful as long as it appreciated that it cannot predicate a single "right" method.
Knowing how a camera sensor really is responding to given amounts of light and how that makes an image look is emphatically a trial and error exercise. There is no empiric standard here either as to what constitutes an acceptable result - it really is a "season to taste" approach.
Good luck with your quest.
Regards
Tony Jay
-
When shooting in liveview against stationary scenes, it ought to be possible for the automatic exposure to make "perfect" exposures in the sense that no sensels are clipped (or at most a user-defined percentage of sensels are clipped) - if artistic choices allows the freedom in selecting aperture and exposure-time (and for some cameras: iso). Anyone know if they do?
-h
-
None that I know of but this would be an advance on current "technology" as far as exposure goes.
Even better would be a real-time RAW-based histogram that would alter accordingly when changes are made to shutter speed, aperture, and ISO.
I bet the means really does exist already but as best I know a commercial expression of it does not.
Regards
Tony Jay
-
Perhaps I'm misunderstanding the original question but there's a are more fundamental reason why determining an 'optimal' ETTR exposure for all conditions isn't possible. That is the dynamic range of the scene itself. If the scene has a narrow brightness range then it may be possible to push the exposure sufficiently that, in the subjective opinion of the photographer, there's enough signal across the entire range to override the evidence of noise. But if you're shooting a scene with a 9 stop range using a camera with a 10 stop sensor it's not going to be possible to do that. You may always have some 'dark' areas where noise will still be too evident; again based on the subjective opinion of the photographer.
-
Perhaps I'm misunderstanding the original question but there's a are more fundamental reason why determining an 'optimal' ETTR exposure for all conditions isn't possible. That is the dynamic range of the scene itself. If the scene has a narrow brightness range then it may be possible to push the exposure sufficiently that, in the subjective opinion of the photographer, there's enough signal across the entire range to override the evidence of noise. But if you're shooting a scene with a 9 stop range using a camera with a 10 stop sensor it's not going to be possible to do that. You may always have some 'dark' areas where noise will still be too evident; again based on the subjective opinion of the photographer.
You are right that automatic ETTR can be difficult when the DR of the scene is large, but in those conditions RAW histograms would be the perfect tool to allow the user to decide:
(http://www.guillermoluijk.com/misc/histogram.gif)
That kind of real time RAW histogram is totally possible, and it allows to find out how much information you would clip with the present shutter/aperture/iso selected (orange bars), and how much information would fall into too noisy areas (dark gray bars). The user would decide which information has to be sacrificed.
The same clipping/noise information could be plotted over the image with blinking areas:
(http://www.guillermoluijk.com/misc/evf_para_alvar.gif)
Regards
-
Oh, I'm not saying camera makers couldn't provide such a tool to photographers, GL. Not at all. All I'm saying is (1) with current technology it's very difficult to determine and (2) if the OP is looking for a formula that can be applied to every image he's looking for too much.
Holy CRAP, GL! We actually agree on something. ;D