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

"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".

Do some research into sound reproduction. Many of the problems are suprisingly similar and sound itself is often represented in graphs logararithmically for the same reason that light is. The actual sound or light would not know that of course, its a deceit to simplify things for the reader and does not disprove the logic here.

Yes, I agree I wanted to know what the sensor can achieve in the real world. I now know that the long tail on the end of my graph is useless for accurate rendering of tonal relationships. I was not looking for anything more.

You seem convinced that the "display" affects the data. I don't see that at all and we could discuss this for ever and not resolve it.

If you can get 12 stops from a 7D sensor, wouldn't you be hearing about this in the reviews? Even Canon's own web site makes no mention of it. I'd like to know where this claim came from.

Lowner wrote in post #10411519
"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".

Do some research into sound reproduction. Many of the problems are suprisingly similar and sound itself is often represented in graphs logararithmically for the same reason that light is. The actual sound or light would not know that of course, its a deceit to simplify things for the reader and does not disprove the logic here.

Yes, I agree I wanted to know what the sensor can achieve in the real world. I now know that the long tail on the end of my graph is useless for accurate rendering of tonal relationships. I was not looking for anything more.

You seem convinced that the "display" affects the data. I don't see that at all and we could discuss this for ever and not resolve it.

No, I am not convinced the display affects the data, not at all, and especially not the raw data. The problem lies with how to make visible (pun intended) all the raw data.

Quotes from DPReview, first 1Ds Mk III:

5D Mk II:

7D:

This is what is possible with colour. If you go B&W, you can go further. The yellowish attached image is the original jpeg, about 7 stops overexposed plus veilign and other artefacts thanks to a cheap filter and the sun reflecting badly off the snow. This was right when I started with dslrs, and not yet saw the need for Raw. Essentially, the complete DR is compressed here in approximately 2 1/2 stops, if that.

HTTP response: 401 | MIME changed to 'text/html'

HTTP response: 401 | MIME changed to 'text/html'

Not sure if anyone has posted the link yet:

http://clarkvision.com​/imagedetail/dynamicra​nge2/

This guy knows more about sensor technology and digital imaging than just about anyone, and I tend to believe his conclusions. One of those conclusions is:
"Digital cameras, like the Canon 1D Mark II, show a huge dynamic range compared to either print or slide film"
Take a look at the shadow detail comparisons to see how much better the shots from the 1D are in terms of retention of detail at the extremes.

wimg wrote in post #10411607
No, I am not convinced the display affects the data, not at all, and especially not the raw data. The problem lies with how to make visible (pun intended) all the raw data.

Quotes from DPReview, first 1Ds Mk III:
5D Mk II:
7D:

This is what is possible with colour. If you go B&W, you can go further. The yellowish attached image is the original jpeg, about 7 stops overexposed plus veilign and other artefacts thanks to a cheap filter and the sun reflecting badly off the snow. This was right when I started with dslrs, and not yet saw the need for Raw. Essentially, the complete DR is compressed here in approximately 2 1/2 stops, if that.



Now, this is what I managed to tweak out of this, recovering most of the detail:

Deepest black is now present, the DR in the picture is approximately 6 stops, and it is no longer overexposed, except the sky.

The picture here doesn't do it any justice, you should really see the 60 cm X 90 cm (24"X 36") print I have on the wall of this picture.

I have another sample somewhere, with a scene DR of 17-18 stops, where the RAW file actually recorded about 13-14 stops, and which I managed to tweak out of it. I'll see if I can find that too.

Kind regards, Wim

Wim, I'd like to see a 100% crop of the bottom left hand corner if possible. It looks very grainy there, on the tree trunk.

There is no question you pull shadows but my original point was that if IQ suffers as much as it seems here then it's not real dynamic range.

This has probably been said here already, but perhaps not in quite the same way...

I think what some are confused by is what exactly the dynamic range of the sensor is. Put simply, it's the ratio of the measured intensity of the brightest light the sensor is able to record before saturating to the dimmest light the sensor is able to record before it becomes undetectable, converted to stops (take the ratio and do a base 2 logarithm operation against it).

You have to measure the actual amount of light involved at each end of the intensity range. That's not something you can really do on the computer once you have the image there. You have to actually perform the measurement at the source. That is, you have to actually measure the intensity of the light the sensor is seeing.

To use any other approach requires great care, because it has to account for the transfer function used to convert the light intensity value into the end value you're making use of. This is especially problematic if the value space being used for the evaluation is smaller than the value space of the sensor itself. Since the value space of the computer display is smaller, the recorded value range of the sensor will of necessity be compressed into the computer display's value space. That compression is lossy, and results in both a dynamic range reduction (to fit the dynamic range of the display) and posterization of the values.

This means you cannot properly interpret the full dynamic range of the sensor without knowing the actual light values it was recording at the time.


Hopefully that makes some sense and is correct...

Wimg,

We are in complete agreement that it is possible to manipulate the basic data. As long as there is data then there will be methods of amplifying it.

Lowner wrote in post #10405302
Film is known to have a better DR, but even that could never achieve 11 or 12 stops.

With NORMAL development, using negative film and a densitometer, you can plot an exposure-density curve over at least 10 and up to 12 stops. You can get this entire density range, certainly up to 10 stops, to print on normal paper if you use variable contrast filters. [newer negative films like T-max have lower dynamic ranges but finer grain than older films]

With PULL development, you can get a dynamic range considerably beyond 12 stops. Bruce Barnbaum and others, using dilute developers with minimal agitation, can get more than 15 stops into a printable range on a piece of film. Sure, the paper it's printed on is not 15 stops, but that's irrelevant. The point is that you can take a single scene with an extreme brightness range of well over 12 stops, expose, develop, and print with no loss of detail in highlights or shadows.

Anecdotally, a friend of mine was shooting IR film but forgot to put on the IR filter. He ended up overexposing a daylit scene by 8 stops, which was his calculated filter factor. He was able to print the negative without any loss of detail.

Sure, it was a dark, dense negative, but there was no highlight detail lost even with an 8 stop overexposure. Try that with any slide film or digital sensor!

DrPablo wrote in post #10412245
With NORMAL development, using negative film and a densitometer, you can plot a film density curve over at least 10 and up to 12 stops. You can get this entire density range, certainly up to 10 stops, to print on normal paper if you use variable contrast filters.

With PULL development, you can get a dynamic range considerably beyond 12 stops. Bruce Barnbaum and others, using dilute developers with minimal agitation, can get more than 15 stops into a printable range on a piece of film. Sure, the paper it's printed on is not 15 stops, but that's irrelevant. The point is that you can take a single scene with an extreme brightness range of well over 12 stops, expose, develop, and print with no loss of detail in highlights or shadows.

Anecdotally, a friend of mine was shooting IR film but forgot to put on the IR filter. He ended up overexposing a daylit scene by 8 stops, which was his calculated filter factor. He was able to print the negative without any loss of detail.

Sure, it was a dark, dense negative, but there was no highlight detail lost even with an 8 stop overexposure. Try that with any slide film or digital sensor!

I don't know the particulars of IR film, but if IR film reactions are limited to radiation in the IR spectrum, then having the filter or not shouldn't have that much of an impact. Your anecdote suggests that IR film would be overexposed due to the addition of visible spectrum radiation. Perhaps it really wasn't. And since we (humans) don't see the IR spectrum, without the use of some kind of IR measuring instrument used on the scene, I find it hard to believe that one could judge any losses or gains of detail.

What I think folks are confusing here are true stops, meaning brightness values which 1/2 as bright as the stop above and twice as bright as the stop below, versus the compression of these stops to tones which are diffierentiated but do not follow the geometrical progression of intensity. Certainly you can compress a scene of 15-20 absolute stops with tones that are within film's ~10 absolute stops to tones that are within photo papers' ~5 absolute stops, but the paper does not have an image of 15-20 absolute stops, but an image with enough gradations as to simulate 15-20 absolute stops.

My original contention was with Lowner's comment that digital sensors only have something like 5 stops (by which I believed he meant absolute stops) which is not true. They have more, and the EV can be measured using a tool such as Rawnalyze on the raw data. Perhaps what Lowner meant was that he could only differentiate tones within 5 absolute stops and tones that are in stops above and below that range appear to be the same as the brightest tone in that 1st stop and the darkest tone in that 5th stop. But you can manipulate those darker and brighter tones like wimg did and push and pull them into the scene such that they are differentiated, but they are not 1/2 or twice as bright.

Poe,

No, thats not what I did. The eye dropper in Photoshop notes the RGB channel brightness between 0 and 255. There is no "visible interpretation" involved. The exposure is adjusted in camera (in my case by 1/3rd stops) so that the eventual series of images goes beyond the images that record 0 or 255. However these "overshoots" can eventually be discarded, they are only there to make sure the limits have been reached. Then the details of the actual exposure are inspected to see what the EV differences are.

The actual test images are not manipulated in any way either.

I don't claim to have invented this rough and ready test. It was the subject of an "Amateur Photographer" article in July 2008.

Lowner wrote in post #10413176
Poe,

No, thats not what I did. The eye dropper in Photoshop notes the RGB channel brightness between 0 and 255. There is no "visible interpretation" involved. The exposure is adjusted in camera (in my case by 1/3rd stops) so that the eventual series of images goes beyond the images that record 0 or 255. However these "overshoots" can eventually be discarded, they are only there to make sure the limits have been reached. Then the details of the actual exposure are inspected to see what the EV differences are.

The actual test images are not manipulated in any way either.

I don't claim to have invented this rough and ready test. It was the subject of an "Amateur Photographer" article in July 2008.

RGB brightness is not, and cannot be used as, an indicator of scene brightness due to the gamma corrections that the RAW data had to go through to be normalized to values between 0 to 255.

It's not limited to the IR. It is sensitive to visible light with sensitivity extended to the near-IR wavelengths.

Poe wrote in post #10412374
I don't know the particulars of IR film, but if IR film reactions are limited to radiation in the IR spectrum, then having the filter or not shouldn't have that much of an impact.

You're waaay off base here. If they were only sensitive to IR you wouldn't need a filter.

IR film is regular old black and white film that has expanded sensitivity into the IR. Some "IR" films are more like "near"-IR, with peak sensitivity at 700-740 nm. Some films have sensitivity well into the IR, like 820 nm. Without a filter it's indistinguishable from regular B&W film.

IR filters, by cutting out all visible light, necessitate increasing your exposure by usually 8-10 stops.

DrPablo wrote in post #10413739
You're waaay off base here. If they were only sensitive to IR you wouldn't need a filter.

IR film is regular old black and white film that has expanded sensitivity into the IR. Some "IR" films are more like "near"-IR, with peak sensitivity at 700-740 nm. Some films have sensitivity well into the IR, like 820 nm. Without a filter it's indistinguishable from regular B&W film.

IR filters, by cutting out all visible light, necessitate increasing your exposure by usually 8-10 stops.

Hence the "i don't know" part.

Poe,

"RGB brightness is not, and cannot be used as, an indicator of scene brightness due to the gamma corrections that the RAW data had to go through to be normalized to values between 0 to 255".

Works for me, and AP.