-SPLIT- Discussion of Dynamic range with some tall claims.

curiousgeorge wrote in post #10406963
I don't think it is.

An image that has a high DR is not automatically contrasty.

Case in point. Velvia is contrasty film, but being slide film it only has 4-5 stops DR. Whereas cheap negative film is 10-12 stops is not necessarily going to be more contrasty.

If the 7D has such high DR it would be able to cope with the first of the OP's images much better, given that it's an overcast sky (if it had been sunny the difference in stops in the scene would have been far greater, making it even harder to capture the full range).

CyberDyneSystems wrote in post #10407064
Indeed quite the opposite,. the higher the DR the capture device is capable of, the lower the contrast will be in the resulting image ( all else being equal )
Unless we are really getting our semantics backwards somewhere along the line.

We can easily do a thought experiment using an example with photo paper:

Lets say this photo paper has a DR of 5, meaning that the darkest tone it can produce when the silver emulsion is exposed to saturation, reflects 6.25% of incident light relative to unexposed emulsion. If you expose half the paper to saturation and prevent any exposure of the other half, develop, fix, etc. you'll have only two tones for your image. Contrasty.

Now lets say we have another paper (with the same white point) with a DR of 10 and do the same half exposed to saturation, half unexposed. This means the dark side of the image reflects 0.2% incident light relative to the white, unexposed portion. If we compare, the paper with DR of 10 has more contrast than the 5-stop paper and has greater DR.

I see your point, the higher the DR the higher the possible contrast as measured in stops, sure,.
But not really, as the highest contrast is still Black vs. White,. and by limiting the DR of your sensor,. you are going to run into those B&W extremes a lot faster than you will with a higher DR sensor.

In practice, the reason we want higher DR is to reduce contrast,.

(using example numbers only) When we expose for the darks,. and the image requires 8 stops to also get the brightest tones without blowout, but we are using a sensor with only five stops,. those bright areas are pure white, blown out.. resulting iamge taken is higher contrast than the real subject matter due to the low DR sensors inability to capture higher DR.

We switch to 10 stop DR sensor, we can capture the full 8 stops, and the brightest areas are not blown out, they still contain tonal data. We have a resulting lower contrast image.

Flip the methodology in how you expose to expose for the brightest areas, and you still get the same effect,. the higher DR sensor will give you a dark tone that is not pure black, the 5 stop sensor will go black too soon. Higher contrast from the 5 stop sensor either way due to it's inability to capture the scene without blowing out the highs or making pure black out of the darks.

The highest contrast we can achieve in an image is pure black and pure white on a page,. but take a photo of two grays that are 10 stops apart with a 5 stop sensor, and you've forced those grays to be B&W... you've increased the contrast with your limited capturing device.
It's the limited DR range that forces images to be captured with higher contrast than the actual subject displays.

Thus higher DR capture device is wanted to reduce contrast.

Back in 2005 this guy http://www.clarkvision​.com/imagedetail/dynam​icrange2/ did some tests leading to the conclusion that a current (at that time) digital sensor had greater dynamic range than print and slide film; presumably digital would be even further ahead today.

stsva wrote in post #10407233
Back in 2005 this guy http://www.clarkvision​.com/imagedetail/dynam​icrange2/ did some tests leading to the conclusion that a current (at that time) digital sensor had greater dynamic range than print and slide film; presumably digital would be even further ahead today.

IMHO, I would not be surprised,. shooting RAW with the 1D mkIII (the first 14 bit Canon DSLR) I felt the DR leverage in post was much improved over the MkII, 1D, 10D, 20D or any other DSLR I'd shot. And certainly better than slide film, if not negative film.

Poe,

"You may consider your 30D as having a "usable" 5 stop range for your photography and processing style, but that does not mean the 30D only has 5 stops of DR"!

I think it does. As I've already said, the long tail at both ends of the DR curve might fool the scientists, but the usable range is nowhere near the range you are quoting. I'll see if I can hunt down the curve I got from my tests, I've been "tidying up" and now cannot find it.

Edited to add: Found it! OK, looking at the graph I could be pushed to admit +3 to -3 is usable, giving me 6 stops if I am prepared to accept the response fall off at the ends. Theres no way on earth that the -7EV is usable, which it needs to be if the 11 to 12 stop range is to be believed.

Has anyone done the same test on a newer body? I'm not interested in "scientific" claims, I'd rather see the results of real world tests. A 450D tested by AP was virtually the same as mine.

By the way, the test procedure:

step1. shoot a mid grey card using "manual" under repeatable and even lighting. Which probably means artificial lighting unless you have more predictable weather than we normally get here in the UK. Use a custom WB to avoid colour messing with the results. Change the shutter speed not the aperture and finish up with a range of exposures showing pure white at one end to pure black at the other.

step2. open each image in Photoshop. Determine its "tone" using the eye dropper. Note down the exposures used that show anything between 0 and 255. Graph the results.

HOSTED PHOTO
please log in to view hosted photos in full size.

CyberDyneSystems wrote in post #10407206
I see your point, the higher the DR the higher the possible contrast as measured in stops, sure,.
But not really, as the highest contrast is still Black vs. White,. and by limiting the DR of your sensor,. you are going to run into those B&W extremes a lot faster than you will with a higher DR sensor.

In practice, the reason we want higher DR is to reduce contrast,.

(using example numbers only) When we expose for the darks,. and the image requires 8 stops to also get the brightest tones without blowout, but we are using a sensor with only five stops,. those bright areas are pure white, blown out.. resulting iamge taken is higher contrast than the real subject matter due to the low DR sensors inability to capture higher DR.

We switch to 10 stop DR sensor, we can capture the full 8 stops, and the brightest areas are not blown out, they still contain tonal data. We have a resulting lower contrast image.

Flip the methodology in how you expose to expose for the brightest areas, and you still get the same effect,. the higher DR sensor will give you a dark tone that is not pure black, the 5 stop sensor will go black too soon. Higher contrast from the 5 stop sensor either way due to it's inability to capture the scene without blowing out the highs or making pure black out of the darks.

The highest contrast we can achieve in an image is pure black and pure white on a page,. but take a photo of two grays that are 10 stops apart with a 5 stop sensor, and you've forced those grays to be B&W... you've increased the contrast with your limited capturing device.
It's the limited DR range that forces images to be captured with higher contrast than the actual subject displays.

Thus higher DR capture device is wanted to reduce contrast.

I think you may be confusing RAW data with an applied tone curve, which correct for black and white points, with the uncorrected linear RAW data, in which case a RAW file from a sensor with greater bit-depth will have greater DR and contrast

Lowner wrote in post #10407281
Poe,

"You may consider your 30D as having a "usable" 5 stop range for your photography and processing style, but that does not mean the 30D only has 5 stops of DR"!

I think it does. As I've already said, the long tail at both ends of the DR curve might fool the scientists, but the usable range is nowhere near the range you are quoting. I'll see if I can hunt down the curve I got from my tests, I've been "tidying up" and now cannot find it.

Edited to add: Found it! OK, looking at the graph I could be pushed to admit +3 to -3 is usable, giving me 6 stops if I am prepared to accept the response fall off at the ends. Theres no way on earth that the -7EV is usable, which it needs to be if the 11 to 12 stop range is to be believed.

Has anyone done the same test on a newer body? I'm not interested in "scientific" claims, I'd rather see the results of real world tests. A 450D tested by AP was virtually the same as mine.

By the way, the test procedure:

step1. shoot a mid grey card using "manual" under repeatable and even lighting. Which probably means artificial lighting unless you have more predictable weather than we normally get here in the UK. Use a custom WB to avoid colour messing with the results. Change the shutter speed not the aperture and finish up with a range of exposures showing pure white at one end to pure black at the other.

step2. open each image in Photoshop. Determine its "tone" using the eye dropper. Note down the exposures used that show anything between 0 and 255. Graph the results.

Your test shows 10-11 stops of DR. Even if you think you only have 5 to 6 usable stops, there is still data that others might find usable.

No, it does not! If you think it does, that explains where the unbelievable claims come from.

Lowner wrote in post #10407591
No, it does not! If you think it does, that explains where the unbelievable claims come from.

Using photoshop or anything else with an eye dropper on a screen for measuring DR is where your thinking is flawed. You are measuring the DR of the screen, not of the RAW file. And it doesn't matter in this case whether you over- or underexpose, because it amounts to the same as photographing a light wedge / grey step card.

A good screen will manage about 8 steps, and if a photograph is processed properly, you will be able to "compress" 12 or even more stops of DR on that screen, no problem. We used to do that with silver emulsion paper, which only had 5 stops. And in order to do that as well as possible, we used something called the Zone System. The only difference is that we now often print to screen, rather than paper, with a slightly higher DR than most old fashioned silver emulsion based papers.

Kind regards, Wim

I don't know, if that graph is your evidence,.

Again, i could be missing something, but isn;t there clearly 9-10 full stops plus of DR?

explain why the -6 and -7 are not usable, and how it would look on the graph if they were usable?
Are you saying that because the "Brightness" axis is down so low, that they are not usable?


I don't understand what the brightness axis of the graph means.
It seems to me that of course the darker stops parts to the left are going to be lower in brightness, so why the two axis graph? If it was a straight line across as opposed to the rising arch, ,. wouldn't that also mean it would only be one stop?

CyberDyneSystems wrote in post #10407664
I don't know, if that graph is your evidence,.

Again, i could be missing something, but isn;t there clearly 9-10 full stops plus of DR?

explain why the -6 and -7 are not usable, and how it would look on the graph if they were usable?
Are you saying that because the "Brightness" axis is down so low, that they are not usable?

I don't understand what the brightness axis of the graph means.
It seems to me that of course the darker stops parts to the left are going to be lower in brightness, so why the two axis graph? If it was a straight line across as opposed to the rising arch, ,. wouldn't that also mean it would only be one stop?

Presumably, because the difference is not really visible on screen anymore?

Kind regards, Wim

Sadly I'm old enough to remember when valve Hi Fi amplifiers where the height of fashion. The very best amplifiers only used the linear portion of the response curve, anything else introduced distortion.

i use the same thinking here, as have AP in the past, although they are slightly more generous with their assessment. Thats why I firmly believe that my 30d has a DR of no more than 6EV. Even AP claims no more than 7.5.

Anything in the long extended tail to the left of the graph is completely distorting any information. Yes, it will record something, but the correct relationship between deep shadows in a scene has been destroyed. The ideal response curve would be virtually straight over much of its length, so I believe an 11 or 12 stop DR graph would look something like this.

HOSTED PHOTO
please log in to view hosted photos in full size.

wimg,

"Using photoshop or anything else with an eye dropper on a screen for measuring DR is where your thinking is flawed. You are measuring the DR of the screen, not of the RAW file. And it doesn't matter in this case whether you over- or underexpose, because it amounts to the same as photographing a light wedge / grey step card".

Perhaps I oversimplified the explanation of the procedure. It does not rely on the quality or otherwise of the monitor screen, only the tonal value as recorded on the sensor. I imagine it could be done using the RAW data in DPP if you feel that Photoshop is not accurate enough. However I wanted a "real world" answer, so the fact that the tools at my disposal give me a certain answer was just what I wanted. I was purposely NOT looking for a laboratory answer.

You are of course right that it is possible to manipulate the data to fit, after all thats what we all do all the time in post processing. But that does not change what we are discussing here one jot.

Lowner wrote in post #10408328
Sadly I'm old enough to remember when valve Hi Fi amplifiers where the height of fashion. The very best amplifiers only used the linear portion of the response curve, anything else introduced distortion.

i use the same thinking here, as have AP in the past, although they are slightly more generous with their assessment. Thats why I firmly believe that my 30d has a DR of no more than 6EV. Even AP claims no more than 7.5.

Anything in the long extended tail to the left of the graph is completely distorting any information. Yes, it will record something, but the correct relationship between deep shadows in a scene has been destroyed. The ideal response curve would be virtually straight over much of its length, so I believe an 11 or 12 stop DR graph would look something like this.

It's different with light values, which is essentially what we are talking about here.

Lowner wrote in post #10408375
wimg,

"Using photoshop or anything else with an eye dropper on a screen for measuring DR is where your thinking is flawed. You are measuring the DR of the screen, not of the RAW file. And it doesn't matter in this case whether you over- or underexpose, because it amounts to the same as photographing a light wedge / grey step card".

Perhaps I oversimplified the explanation of the procedure. It does not rely on the quality or otherwise of the monitor screen, only the tonal value as recorded on the sensor. I imagine it could be done using the RAW data in DPP if you feel that Photoshop is not accurate enough. However I wanted a "real world" answer, so the fact that the tools at my disposal give me a certain answer was just what I wanted. I was purposely NOT looking for a laboratory answer.

You are of course right that it is possible to manipulate the data to fit, after all thats what we all do all the time in post processing. But that does not change what we are discussing here one jot.

I respectfully disagree.

Even when working from Raw, what you are measuring, whatever you do, unless you really measure directly from the sensor, does not make sense in order to determine what the DR is, unless, and I say unless deliberately, you manage to fit that what the sensor reads in a way that is possible to display fully on a screen. However, on a screen you will never go beyond 8 stops or thereabouts, because that is at best what a screen is capable off.

If you really want to find out what the true DR of a sensor is, there is only 1 way, namely photographing a grey stepping chart with all the possible grey steps for DR you want to investigate, expose for middle grey (18 % grey), then "develop" the image in such a way that you can see as many as possible steps of the chart, IOW, translate that which the camera sensor can see to that which we can see on screen.

You could compare this, going to your example, to an audio system with a frequency range from 10 to 100,000 Hz within the THD you want to achieve, even a linear response if you like, by means of using your ears, IOW, a "display" device that can only handle about 20 - 20,000 Hz at best.

Unless you use some kind of system to "display" (hear) the 10-20 Hz bit and the 20,000 to 100,000 Hz bit, how are you ever going to prove it is capable of that?

Essentially, this last step is what you are skipping with your method. The simple way to make this part of the range visible is by compressing the grey steps so that they will fit within the range used to display them, i.e., the DR your screen is capable of or the DR your print is capable of, or you need specialist measuring tools to read out data directly from the sensor, as Raw has no direct, or 1:1 relation with DR either. It already is a translation. Do note that when you measure this output, it will only have a maximum of about 8 stops of DR, even idf the real recorded stuff has more.

BTW, have a look at some of the older tests at DPReview where they do check the maximum DR of a camera sensor. This does correspond to my own findings, for example when doing a single shot HDR type processing, by creating two files from the same raw file, one with the low light values, and one with the high light values. Just doing this already proves there is way more than 5-6 EV.

If you actually don't mind some noise, which is not visible at about 8"X 12" it is even possible to extract about 12 -15 stops of DR from a 5D sensor. And yes, I do have to use Zone techniques in order to display those, and work in 16 or even 32 bit files during the process, for optimal working range and lose as little information as possible.

That these values aren't necessarily linear, doesn't really matter, as long as you can distinguish these with the naked eye. Human sight isn't linear either. It's only more or less linear in average light, in the middle area in the curve other words.

Kind regards, Wim

"It's different with light values, which is essentially what we are talking about here".

No, I see absolutely no difference, it's a simple linear response to the scene, a one to one relationship (or should be in a perfect world). Currently the technology is not capable of that but we are still in the very early days of the "digital revolution", come back in 100 years time! This kind of response curve is not unique to photography, its common to most electronic/electrical devices which is what we are discussing and as I said, the valve amplifier solution was to flatten the centre of the curve by clever design then simply not use that part of the curve that is non linear.

"If you really want to find out what the true DR of a sensor is, there is only 1 way, namely photographing a grey stepping chart with all the possible grey steps for DR you want to investigate, expose for middle grey (18 % grey), then "develop" the image in such a way that you can see as many as possible steps of the chart, IOW, translate that which the camera sensor can see to that which we can see on screen".

If you manipulate the data in post-processing to bring it back to "mid grey", then you are doing exactly what you criticised me for and getting similar results. I see no gain in shooting a complete grey scale when I only need a single reference sample? It will be impossible to retain the extremes of the greyscale wedge, however you might come up with some interesting results. Care to try it and show us the results?

You mentioned the "display" again. Non of the measuring relies on the monitor in any way. We are using information obtained directly from the image, what that actually looks like is of interest of course but the results would be the same whether I used the worst monitor ever made or the best. Yes I am the first to agree that my rough and ready approach is not "scientific" and thats because I wanted a "warts and all" honest answer to the question "whats the practical dynamic range of my 30D" - The answer is 5 stops but it can be pushed or pulled to give me a little more if I really need to force things. Thats what I was after and using spot metering as I do most of the time I use the information gained constantly.

Hi Richard,

Lowner wrote in post #10411443
"It's different with light values, which is essentially what we are talking about here".

No, I see absolutely no difference, it's a simple linear response to the scene, a one to one relationship (or should be in a perfect world).

I still don't understand why? You can't show a 1:1 relationship, not on any medium I know about. And the relationship is not linear anway, it is exponential, or logarithmic.

No, it isn't. The difference is that I am trying to display the whole DR in one shot on a medium (print or screen) that ísn't capable of showing the same full DR. The way I see it, what you do is showing the limits of the medium, not the sensor. The only way to show or see the "linearity" and complete DR of a sensor, is to "print" a picture of something with, let's say, a DR range of 20 stops (or more), and then "compress" the recording (the Raw file) in such a way that in the output all is visible in one go. In order to be linear, the different DR steps still visible, should have a logarithmic or exponential nature (when nioise is subtracted). If you just shoot a picture of a single grey value, that is not going to work, because you don't know the relation of the different outputs to each other.

Part of your reply seems to have fallen away, unfortunately.

If I find some time, I'll see if I can find some real life examples.

Kind regards, Wim