Home · Register · Search · View Winners · Software · Hosting · Software · Join Upload & Sell

Moderated by: Fred Miranda
Username   Password

  New fredmiranda.com Mobile Site
  New Feature: SMS Notification alert
  New Feature: Buy & Sell Watchlist
  

FM Forums | Alternative Gear & Lenses | Join Upload & Sell

1
       2       3       4       end
  

Archive 2013 · Dynamic Range vs. Tonal Range
  
 
RustyBug
Offline
• • • • • •
Upload & Sell: On
p.1 #1 · p.1 #1 · Dynamic Range vs. Tonal Range


I know this is more of a general question than an "alt question", but I have significant respect/trust @ this board for technical issues ... and my question stems from comparing newer alt cameras against older non-alt cameras.

I'm looking at DXO measurements (take it for what its worth) and noticing that the three cameras that I've chosen to compare are like "splitting hairs" on nearly all aspects, except for DR. The three cameras come from three different platforms 4:3 (OMD E-M5), APS-C (Pentax K5 IIS), and APS-H (Canon 1D MK II N).

While the DR difference didn't seem that strange given the age variance, the SNR 18%, Tonal Range and Color Sensitivity all went to the APS-H, 2005 model (albeit only slightly). So, my question is this:

What is the primary driver(s) at why SNR 18%, Tonal Range & Color Sensitity would be so similar (edge to the 1D II N) and yet the DR be so significantly different? And, how would the use of polarized / ND filters impact the DR (exposure impact obvious) of the different platforms?



Apr 17, 2013 at 02:55 AM
kwalsh
Offline
• • •
Upload & Sell: Off
p.1 #2 · p.1 #2 · Dynamic Range vs. Tonal Range


SNR 18% is entirely a function of how many photons the sensor counts at a given ISO. The only two things that affect that are sensor size and quantum efficiency. QE hasn't actually changed much over the years hence everything looks very similar and larger sensors tend to do better. It is essentially a limit of physics (photon counting statistics) more than sensor technology these days.

The way DxO measures Tonal Range and Color Sensitivity tends to closely relate to SNR 18%.

DR, as measured by DxO, relates to the well depth at base ISO and the read noise at base ISO. These two parameters are all about sensor design and there have been big changes in the past few years that improve it significantly - specifically the read noise part.



Apr 17, 2013 at 03:21 AM
RustyBug
Offline
• • • • • •
Upload & Sell: On
p.1 #3 · p.1 #3 · Dynamic Range vs. Tonal Range


Thanks.

"read noise" ... is that relative to jpg, raw or both?



Apr 17, 2013 at 03:34 AM
kwalsh
Offline
• • •
Upload & Sell: Off
p.1 #4 · p.1 #4 · Dynamic Range vs. Tonal Range


All the DxO measurements are based on RAW. Once you go to JPEG all bets are off as noise reduction is applied by the camera.

Read noise is essentially what you would see in the RAW file if you took a short exposure with the lens cap on. You would expect to see black, but instead you see a bit of noise. That's the read noise, noise in the circuitry that reads the pixels. This is different from say the noise in a blue sky. The read noise is still there, but it isn't significant as there is much stronger "shot noise" which comes from the statistics of photon counting. Once you have any significant exposure beyond the shadows the noise is dominated by shot noise.

So SNR 18% is typically a measure of shot noise, which is entirely a function of how many photons hit the sensor - so for a given ISO proportional to the sensor size. DR is typically actually measuring read noise, which is a function of the design of the sensor circuitry. Over simplification but that's the general idea.



Apr 17, 2013 at 03:48 AM
philip_pj
Offline
• • • •
Upload & Sell: Off
p.1 #5 · p.1 #5 · Dynamic Range vs. Tonal Range


Here is a take from Tim Ashley on DxO sensor metrics:

'Nonetheless, I have never seen a DXO Sensor score for a camera with which I have experience that disagrees with my own 'unmeasured' feeling for the qualities of that sensor. I therefore consider it a very useful guide, for cameras I am thinking of buying, as to whether or not the imaging performance of their sensors will suit me or not.'

I agree and use them freely to get an idea of what is the lowdown. Not so much for the headline numbers as for the details and patterns, for example one of my cameras has two stops of DR over my a900, but loses almost 6 stops from base ISO to ISO 6400. And you do see the effects, lots of gritty street work is very high contrast.

http://tashley1.zenfolio.com/blog/2013/3/whats-your-doh-mark-score

DxO - their Tonal range definition:
'Tonal range is the effective number of distinguishable gray levels a camera can produce.'

It follows a simialr pattern to DR, not surprising when you think about what it is measuring. I think both measures assist colour handling, makes it less 'posterised', more subtle gradations.



Apr 17, 2013 at 10:14 AM
RustyBug
Offline
• • • • • •
Upload & Sell: On
p.1 #6 · p.1 #6 · Dynamic Range vs. Tonal Range


Thanks guys, that is helping a bit.

I've googled "read noise" to find this link:
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/

I followed about 1/2 of it reasonably well. Having a better understanding how to offset/correct for it in practical terms for shooting strategy and pp strategy would be helpful. I've struggled a bit with noise, so I figure I should get a better grasp of it ... before I make any new camera purchases to augment (i.e. not replace) my current gear.

BTW, interesting link re: DXOMark ... it basically parrots my perspective of DXOMark ... and says it in a way I never could.



Apr 17, 2013 at 11:39 AM
kwalsh
Offline
• • •
Upload & Sell: Off
p.1 #7 · p.1 #7 · Dynamic Range vs. Tonal Range


As for shooting strategy the only thing that really helps noise is capturing more photons - i.e. more exposure over a larger sensor. Shoot at the lowest ISO possible. Once at base ISO you can improve things a bit more by "exposing to the right". You can improve things even more by doing AEB at base ISO and then combining the exposures to an HDR file (note, I don't mean necessarily doing HDR tone-mapping - just creating an HDR file which in recent versions of LR you can process directly without tone mapping, PhotoAcute also has a similar feature to create a DNG file merged from multiple exposures).

Obviously the above only works on certain subjects. For things that move (either the subject or the camera) you'll be limited by motion blur and that will eventually force you away from base ISO or HDR. Faster lens, bigger sensor, newer sensor are your only options at this point for capture.

As for PP it is all about balancing noise reduction and sharpening. The LR/ACR tools are quite good these days but I know some of the dedicated NR software like Noise Ninja still do a bit better.

Anything that increases local contrast (contrast, steeper tone curves, clarity, saturation, vibrance, sharpening) will amplify noise. If you can be selective about such adjustments (e.g. don't sharpen out of focus areas) that might help.

Finally, consider your shadows. Do you actually need detail there? Noise is worst in the shadows so consider the trade off, pushing the shadows to get detail there but also ugly noise or letting the shadows block to a solid black. This is a composition question. Technique and technology freaks will often fight for shadow detail without considering whether the image is actually any better with that detail. Painting is an additive process, the artist must add each element to the composition. Photography is a subtractive process, the artist actively must remove elements from the composition (be it tighter framing, selective focus, dodging/burning) to keep the viewers eye focused where the photographer wants it. So don't blindly amp up the shadows, only do so if what is in the shadows adds to the composition.



Apr 17, 2013 at 12:53 PM
RustyBug
Offline
• • • • • •
Upload & Sell: On
p.1 #8 · p.1 #8 · Dynamic Range vs. Tonal Range


Thanks.

I shoot mostly FF @ base ISO, so I'm pretty good in that regard.

I guess the thing that has raised the question here for me is that if the tonal range/SNR/color sensitivity is essentially the same for sensor A vs. sensor B vs. sensor C ... yet the dynamic range is different ... what does that mean for real world shooting?

In other words, if I'm getting the tonal range/SNR/color sensitivity either way (even slightly better with the lower DR sensor) ... how is the DR difference hurting/helping/limiting me in theory vs. practical terms (@ raw captures).






Apr 17, 2013 at 01:32 PM
kwalsh
Offline
• • •
Upload & Sell: Off
p.1 #9 · p.1 #9 · Dynamic Range vs. Tonal Range


The way "DR" is measured on DxO what it really is telling you is what the noise in the shadows looks like.

So for real world shooting, if you put a lot of positive onto the "Shadows" slider in LR/ACR (or "Fill Light" in other converters) and end up with noisy pushed shadows then a camera with a better "DR" measure on DxO might help you.

If on the other hand you rarely have shadows problems but instead don't like noise in the mid tones (e.g. noise in blue skies) then a new body with a higher DR measurement is unlikely to help you out at all.

In the mean time, if you discover you are rarely using much negative on the "Highlights" or "Whites" slider in LR/ACR, are rarely doing negative exposure in post processing and rarely if ever find blown unrecoverable highlights in your shots then the other thing you could try besides a new body would be to dial in some positive exposure compensation (expose to the right a bit). This is effectively like creating a lower ISO setting for your existing camera, noise will be better in both the shadows and the mid-tones but the trade off is that you run a greater risk of blown highlights. You haven't changed the DR of the camera at all, you've just shifted where you expose so that you use more of the highlight range (the lowest noise part) in your images.

I see from your profile you shoot Canon. Base ISO DR is not something Canon sensors are great at, the newer Sony sensors are pretty amazing in this regard. Not too much improvement in Canon land over the past five years in this regard. So I honestly don't think a new Canon body is going to give you much DR improvement.

P.S. If you shoot static subjects, landscape for example, give AEB a try. In PP pick the most exposed shot that hasn't blown the highlights (i.e. for each shot in the AEB set counter act the in camera exposure compensation with an equal one in PP, now all the exposures look the same except the shorter ones have more noise and the longest ones have clipped highlights). This is much easier than "exposing to the right" while shooting where you have to think and risk blown highlights. You are essentially "exposing to the right" in post processing by picking the best RAW file to work with.



Apr 17, 2013 at 01:54 PM
RustyBug
Offline
• • • • • •
Upload & Sell: On
p.1 #10 · p.1 #10 · Dynamic Range vs. Tonal Range


kwalsh wrote:
The way "DR" is measured on DxO what it really is telling you is what the noise in the shadows looks like.

So for real world shooting, if you put a lot of positive onto the "Shadows" slider in LR/ACR (or "Fill Light" in other converters) and end up with noisy pushed shadows then a camera with a better "DR" measure on DxO might help you.

If on the other hand you rarely have shadows problems but instead don't like noise in the mid tones (e.g. noise in blue skies) then a new body with a higher DR measurement is unlikely to
...Show more

Thanks again.

I shoot mostly on my Kodak SLR/c for manual focus alt glass, base ISO, and use my Canon for AF/fps needs ... using the EF mount for versatility @ alt glass on FF / APS-H as my platform. Old school, I shot Fujichrome 50 (fine grain) and knew what a tripod was all about ... not recalling any issues other than reciprocity failure compensation/color shift ... but grain (noise) was not an issue for me. Trying to get back to that somewhat.

In digital terms, I struggle a bit at grasping the technicals / strategies on the the low end of the spectrum with how to approach longer exposures ... well depth, noise, etc. It sounds like you are suggestion overexpose the shadows @ capture, then pull them back in pp ... blending with a properly exposed image in midtones/highlights.



Apr 17, 2013 at 03:06 PM
 

Search in Used Dept. 



buggz2k
Offline
• • •
Upload & Sell: Off
p.1 #11 · p.1 #11 · Dynamic Range vs. Tonal Range


Good reading, thanks for the infos!


Apr 17, 2013 at 03:20 PM
Alan321
Offline
• • • • •
Upload & Sell: Off
p.1 #12 · p.1 #12 · Dynamic Range vs. Tonal Range


RustyBug wrote:
In digital terms, I struggle a bit at grasping the technicals / strategies on the the low end of the spectrum with how to approach longer exposures ... well depth, noise, etc. It sounds like you are suggestion overexpose the shadows @ capture, then pull them back in pp ... blending with a properly exposed image in midtones/highlights.


Yep. That's what ETTR (expose to the right) is all about. It should be expose to the right and process to the left, because unless you correct the overexposure your images will always be overexposed. However, the problem with ETTR is that you must not go so far that important highlights are clipped at the bright end of the scale. And that clipping means clipping in any of the individual colour channels. e.g. you can clip red for a bright red flower and yet the luminance may still be quite dark.

In practise, ETTR is complicated by the fact that the histogram is based on jpeg pixel data rather than the captured raw pixel data. The jpeg pixel data (and histogram) is affected by a bunch of camera settings such as contrast, saturation, picture style, sharpness, white balance, etc., in addition to the exposure settings that affect the raw pixel data.

If you want to capture clean shadows / dark areas and unclipped bright highlight details then you want to use a camera / ISO combination with lots of DR, or take multiple shots at different exposures and combine them in post processing.

DxO like many others defines the dark limit of the DR range as the level at which noise = signal. I reckon that is too optimistic because of half of my pixels were just noise then I'd say it was a noisy picture. I reckon a more useful bottom limit is more like where signal is 2 or 3 stops brighter than noise level, because then the subject would be easily distinguished from the noise.


Just to show that numbers can be manipulated to look great and yet the camera be useless, consider this theoretical example: Imagine that a camera was set up to produce only three data levels corresponding to 0 at pure black, 128 at middle tone and 255 at pure white. The DR is then a perfectly clean 8 stops (for 8-bit data) but the tonal range comprises just 3 discrete levels - not even enough for a good mono image. You would expect to see noise free skies but they would be horribly posterised.


It is easy to consider that dark shadows are often irrelevant but not in high DR landscapes in which much of a forest undergrowth is relevant but poorly lit while there may also be brightly lit areas of interest. Also, what many forget is that dark subjects in bright light can still be dark and noisy. e.g. a black car in sunshine should not be allowed to look noisy just because it is painted black. In both examples ETTR will likely not be applicable because of relevant bright areas that must not be overexposed.

- Alan



Apr 19, 2013 at 01:26 PM
RustyBug
Offline
• • • • • •
Upload & Sell: On
p.1 #13 · p.1 #13 · Dynamic Range vs. Tonal Range


Alan321 wrote:
It should be expose to the right and process to the left


that missing piece of information just might be kinda significant, thanks.
Basically pull/push or push/pull (I forget how that went ... I shot chrome, not negs.)



+1 @
And that clipping means clipping in any of the individual colour channels. e.g. you can clip red for a bright red flower and yet the luminance may still be quite dark.

In practise, ETTR is complicated by the fact that the histogram is based on jpeg pixel data rather than the captured raw pixel data. The jpeg pixel data (and histogram) is affected by a bunch of camera settings such as contrast, saturation, picture style, sharpness, white balance, etc., in addition to the exposure settings that affect the raw pixel data.


Realizing that your histo is not telling you the truth regarding your raw ... strategy for assessing your capture in the field?



If you want to capture clean shadows / dark areas and unclipped bright highlight details then you want to use a camera / ISO combination with lots of DR, or take multiple shots at different exposures and combine them in post processing.

I shoot mostly base ISO.



Just to show that numbers can be manipulated to look great and yet the camera be useless, consider this theoretical example: Imagine that a camera was set up to produce only three data levels corresponding to 0 at pure black, 128 at middle tone and 255 at pure white. The DR is then a perfectly clean 8 stops (for 8-bit data) but the tonal range comprises just 3 discrete levels - not even enough for a good mono image. You would expect to see noise free skies but they would be horribly posterised.

+1 @ example ... got it @ diff of DR vs. TR.
so if 0-255 = 8 stops ... how do we get cameras that have a DR of more than 8 stops



Apr 19, 2013 at 04:12 PM
AhamB
Offline
• • • • •
Upload & Sell: On
p.1 #14 · p.1 #14 · Dynamic Range vs. Tonal Range


RustyBug wrote:
Realizing that your histo is not telling you the truth regarding your raw ... strategy for assessing your capture in the field?


Set the jpeg settings to the most neutral profile/least contrasty setting and use UniWB. UniWB helps a lot to know whether you're blowing out color channels or not (handy when shooting brightly colored things such as flowers). Any camera that allows you to set a custom WB (selected from a RAW file) can use UniWB with this method: http://www.guillermoluijk.com/tutorial/uniwb/index_en.htm


RustyBug wrote:
+1 @ example ... got it @ diff of DR vs. TR.
so if 0-255 = 8 stops ... how do we get cameras that have a DR of more than 8 stops

Most cameras have 12 or 14 bits, remember? 2^8 = 256 levels; 2^12 = 4096 levels; 2^14 = 16384 levels. I don't remember how that exactly translates to DR, but I guess it has something to do with having enough levels to differentiate signal from noise (esp. in the lower stops because they get awfully few levels if you only have 256 in total). Question for the experts: what was the reason again that the lower stops have less levels? Is that true for the linear sensor data or only after applying the gamma curve?

Edit: I guess it's just how binary counting of the well charge works... you use 12 or 14 bits to count the maximum signal (full well charge), so that means the lowest possible signal can only be represented by the first bit, which is only one level (0 or 1). I'm still not sure how the stops/exposure values are distributed over the bits though. Is it because bit count(2^x) is the inverse function of exposure (log2?)?



Apr 19, 2013 at 04:55 PM
RustyBug
Offline
• • • • • •
Upload & Sell: On
p.1 #15 · p.1 #15 · Dynamic Range vs. Tonal Range


But, if black = 0 and white = 255 ... i.e. 8 stops, aren't the extra levels really only additional degrees of refinement (TR ) within the range of 0-255. Sure, that means that the smallest near black @ 1 vs. 0 is not as refined as say .01 (not sure what the 14 bit equivalent would be) vs. 0 ... but it's not like we are actually getting anything beyond the 0-255 range (i.e. more dynamic range ), are we?

I can see how say from 0-30 in a 14 bit realm would have smoother tonal transitions that 0-30 in an 8 bit realm ... but how does the change in bits, increase dynamic range other than the ability to have a smaller denominator (i.e. less than 1) to produce a larger quotient?

Or am I way off base here



Apr 19, 2013 at 06:05 PM
Kolor-Pikker
Offline
• •
Upload & Sell: Off
p.1 #16 · p.1 #16 · Dynamic Range vs. Tonal Range


AhamB wrote:
Most cameras have 12 or 14 bits, remember? 2^8 = 256 levels; 2^12 = 4096 levels; 2^14 = 16384 levels. I don't remember how that exactly translates to DR, but I guess it has something to do with having enough levels to differentiate signal from noise (esp. in the lower stops because they get awfully few levels if you only have 256 in total). Question for the experts: what was the reason again that the lower stops have less levels? Is that true for the linear sensor data or only after applying the gamma curve?

Edit: I guess it's just how
...Show more

Since sensors are by nature linear, the number of bits is directly proportional to how much full stops of dynamic range the electronics can resolve, assuming the sensor is good enough to provide that many bits of data.
Up until now, most AD converters were 14-bit because there are only a handful of cameras that can actually reach 14 stops of DR. It is actually possible for a sensor to have greater DR than the bit depth of the electronics.

The distribution of stops over the exposure range is identical for every digital sensor out there, with the brightest stop consuming half of all available values, and each consecutive stop taking up half as much as the last one.

Naturally, if you were to interpret this directly, you'd get a very bland image, and you can by setting some RAW converters to use "linear" gamma. Lightroom doesn't allow you to change this directly in the software, but you can use DNG profile editor to make your own Linear profile to use in LR. The reason for the blandness is as explained, the highlights take up the right half of the histogram, with the mid-tones and shadows bunched up to the left. To remedy this, RAW converters use what's called a gamma or "film" curve to vaguely simulate the way light is recorded on film, because that's what we are used to seeing, and does so by compressing the highlights and lifting the mid-tones to where we expect.



Apr 19, 2013 at 06:28 PM
AhamB
Offline
• • • • •
Upload & Sell: On
p.1 #17 · p.1 #17 · Dynamic Range vs. Tonal Range


I hope one of our experts can answer that. Michael/mpmendenhall is writing something that seems to be relevant here: http://www.fredmiranda.com/forum/topic/1200319/6#11452988

Using less bits than the DR needs would obviously lead to clipping/posterization. I guess that doesn't have anything to do with noise, but just lack of tonal resolution. If you protect the highlights, the posterization in the shadows limits the DR because it just turns to solid black I guess.

Edited on Apr 19, 2013 at 06:53 PM · View previous versions



Apr 19, 2013 at 06:48 PM
kwalsh
Offline
• • •
Upload & Sell: Off
p.1 #18 · p.1 #18 · Dynamic Range vs. Tonal Range


I'll just interject quickly here (don't have much time at the moment) to be extremely careful when bringing up the concept of "levels" or "bits" in the RAW file as relates to either DR or TR. It turns out in essentially every camera ever made the bit depth of the ADC or the RAW file has been irrelevant to DR or TR. The designers, not being idiots, have always ensured adequate bit depth in the digital data such that the sensor is limiting, not the ADC or RAW file resolution. And in fact in many cameras that have both a 14-bit and 12-bit mode the 14-bit mode does nothing at all to improve either DR or TR because the sensor isn't good enough to need 14-bits (however the most recent Sony sensors can actually make use of 14-bits at the lowest ISOs).

ETTR, and tonal range are all about noise - not bit depth of the RAW file (assuming the designer is not daft). In the deepest shadows the noise limit is "read noise". In all other tones it is "shot noise". Were someone to make a "perfect" sensor with no "read noise" there would still be a DR limit imposed by "shot noise" in the shadows.

Sorry I don't have time to write more, but some Googling should sort most things out.

Ken



Apr 19, 2013 at 06:52 PM
RustyBug
Offline
• • • • • •
Upload & Sell: On
p.1 #19 · p.1 #19 · Dynamic Range vs. Tonal Range


Thanks Ken (et al) ... then that would kinda explain why a 2005 model sensor would yield same (slightly better) than the newest, latest, greatest, highest DR @ SNR, TR, CS, despite "lagging" in the DR measurement department ... depending on whether we are talking about read noise vs. shot noise, etc.

I think I "kinda" got it now ... will have to re-read through this thread and combine with the other one, but I think the pieces are mostly there for me now.

Thanks all.



Apr 19, 2013 at 07:41 PM
theSuede
Offline
• • • •
Upload & Sell: Off
p.1 #20 · p.1 #20 · Dynamic Range vs. Tonal Range


Rusty, do mind to ALWAYS make a clear distinction between the "object" and "image" sides! Otherwise confusion will just continue to grow...

One ("object" of course...) is in front of the lens. This is the reality side, where DR (can be) great, and both tone (grade) and spatial (detail) resolution approaches infinity at long integration times (very, very low ISO values in capture). This is the basis for the input DR, you measure what input range in front of the camera in which the camera can accurately still resolve detail.

The other, the "image" side, is the 2-D image that the lens projects on to the sensor or film for the medium to capture. Hopefully the lens lets through at least 11-12Ev's of very good detail material in situations with no or little flare in the glass. At about 15Ev, you're getting severely impeded by random scatter in the glass and the glass surfaces - what we call "veiling flare". The "Image" can be represented by any method or scale that you chose, as long as you KNOW how conversion between input>output was done, you can get back to the original object value.

The image side has to be RECORDED. In practice it has to be translated into some value range that gives you enough tone resolution without taking up to much data space. Raw data is almost always linear, that is - if you double the exposure, you double the raw value. 1-2-4-8-16-32-64-128-256-512-1024-2048-4096-8192-16384 - those are the 14 full steps in a 14-bit raw file. Note that EV3 goes from 4 to 8, that is 4 discrete "steps" of tone resolution. Look further up, at EV11 that goes from 2048 to 4096 - that's 2048 steps of tone resolution in one Ev!

Now, if you convert that by using a gamma of "2.0", close to standard, into the discrete levels of 1=1, 2=2, 3=4, 4=8, 5=16, 6=32, 7=64, then you only need 14 bit steps - that 4 bits! - to cover all 14 Ev's of input! But your tone resolution will be atrocious, you'll have serious posterizing. Only 14 gray scale levels.

But! Now those four bits actually "cover" 14Ev of the original object DR you put in to the system... 4bit = 14Ev DR in this case!

jpg uses 8 bits, 256 discrete levels, which is enough to get rid of almost all posterizing - but it doesn't leave much room for added pulling or pushing without making the discrete steps visible.



Apr 20, 2013 at 02:02 PM
1
       2       3       4       end




FM Forums | Alternative Gear & Lenses | Join Upload & Sell

1
       2       3       4       end
    
 

You are not logged in. Login or Register

Username   Password    Retrive password