Dynamic range of some model cameras

Apparently, there is no standard criteria for measuring dynamic range on a digital camera.

Tom W wrote in post #3652760
Apparently, there is no standard criteria for measuring dynamic range on a digital camera.

I bet camera makers love that fact: they can try pushing their DR numbers up

Dynamic range has nothing to do with JPG or RAW, it is a limitation of the sensor. The sensor captures the RAW image which can then be converted to JPG, with all the dynamic range intact. What is lost is the number of brightness values between the endpoints of that range.

If the full 8+ stop range is used it is very likely there will be some gaps in the JPG. The RAW, with 16 tmes as many values will do much better at keeping all the values.


boB

davesrose wrote in post #3651984
JPEG is like photographic paper: it's 8bpc and can not have much tonal range past 4 stops.

Wait, are we talking about tonal range or dynamic range?

boBquincy wrote in post #3654692
Dynamic range has nothing to do with JPG or RAW, it is a limitation of the sensor. The sensor captures the RAW image which can then be converted to JPG, with all the dynamic range intact. What is lost is the number of brightness values between the endpoints of that range.

If the full 8+ stop range is used it is very likely there will be some gaps in the JPG. The RAW, with 16 tmes as many values will do much better at keeping all the values.

boB

While it is possible to convert a RAW to JPG with the full number of stops of information intact, in practice, that doesn't happen. You will lose data as the camera or software struggles to pull a roughly 7-8 stop image out of the data provided by the camera. In my experience, a good part of that lost data occurs in the highlights, which seem to be clipped off on the JPG (perhaps in an effort to maintain some reasonably smooth gradation at the lower light levels or zones - something has to give).

You can cover the full spectrum of brightness from pure black to pure white in 1 or 2 bits if you want, with 0 or 00 being black and 1 or 11 being white. But the results would be rather horrible in terms of image quality.

A couple of years ago, I did some experimenting with my 1D II, comparing the ability of the camera to recover highlights when shot in RAW vs. JPG. Here's the results:

The original image, properly exposed:

kpt4321 wrote in post #3654742
Wait, are we talking about tonal range or dynamic range?

both....they are related: tonal range is the value range of the colors that are in your photograph. Dynamic range is the value range of luminosity (or reflected light) that's in the scene. Film and digital sensors have more dynamic range then ink or photo paper: so they give you more adjustment with the tonal range of an image. I like to think about tonal range of an image when I'm PP, and dynamic range when I'm shooting: just as Adam's zone system does for camera vs print.

boBquincy wrote in post #3654692
What is lost is the number of brightness values between the endpoints of that range.

If the full 8+ stop range is used it is very likely there will be some gaps in the JPG. The RAW, with 16 tmes as many values will do much better at keeping all the values.

Dynamic range is effected by luminosity values....especially with 8bpc. Lets say the dynamic range of the sensor is 10 stops. Yes, RAW and Jpeg might keep those 10 stops....but RAW has 4092 levels of luminosity vs Jpeg's 256. Now exposure is logarithmic, so your brightest stop gets half of your levels: 2046 levels, jpeg: 128 levels. So now you have 2046 levels in a 12bpc system to be distributed to your next 9 stops......128 levels to your next 9 stops in jpeg. Keeping in mind that the luminosity levels are going to be higher in the brighter stops, you can see that in a 8bpc system, you really don't have enough luminosity levels to fill the darkest stops in a 10 stop range.

In Tom's example above, we can see how a more limited number of luminosity levels effects dynamic range and midtones of an image. In a 8bpc system, every ounce of tone has to be distributed evenly across the image. Suddenly, when you try to adjust its curves, you'll lose detail in your midtones: as there was not enough value range in those areas to begin with (4092 shades of distribution vs 256 is nothing minor). Even though the MarkIII may just have a few more luminosity levels then the previous generation Canons, it's still an improvement. And to say it again, I can't wait until we have 32bpc luminosity in our cameras!!!!

In that case, would you please prove this statement?

davesrose wrote in post #3651984
JPEG is like photographic paper: it's 8bpc and can not have much tonal range past 4 stops.

kpt4321 wrote in post #3659672
In that case, would you please prove this statement?

Tonal range of an image can only be manipulated within the dynamic range of the image file itself. In a classical RGB color space, that's 256 shades of tone. Jpeg has a total of 256 luminosity levels to distribute....12bpc RAW has a total of 4092. Tom's examples are proof that jpeg has more limited tonal range then RAW....and here's a bit more:

Here are the terms for the different DRs of photography:

http://en.wikipedia.or​g/wiki/Exposure_range

I just found this article which describes the differences in contrast range and what that means in F stops between the different bit depths:

http://www.cambridgein​colour.com/tutorials/d​ynamic-range.htm

Tom W wrote in post #3654794
......A couple of years ago, I did some experimenting with my 1D II, comparing the ability of the camera to recover highlights when shot in RAW vs. JPG. Here's the results:

The original image, properly exposed:

SEE IMAGE IN POST BY TOM W

Overexposed by 2 stops:

SEE IMAGE IN POST BY TOM W

Now, I took a RAW image that was overexposed by 2 stops and pulled it down in Canon DIgital Photo Pro software. Converted to JPG for internet purposes, here is the result:

SEE IMAGE IN POST BY TOM W

Shooting the same overexposed image in JPG and then trying to pull the exposure down in post-processing resulted in this:

SEE IMAGE IN POST BY TOM W

As can be seen, the RAW image presented useful data at an exposure level considerably higher than the JPG did. The JPG's overexposed image just couldn't be recovered with any reasonable fidelity. My conclusion was that I was gaining at least a stop of useful data at the bright end of the scale by using RAW.

Not sure that can be considered an increase in dynamic range by definition, but it's certainly an increase in useful dynamic range. That is, there's data in the RAW image that can be used, but which is missing in the JPG.

I have been following this thread and have found it to be very interesting and informative. I certainly agree with your statement in the last paragraph that increasing the number of bits does not increase the dymanic range -- it just improves the fidelity of the shades of luminosity between black and white -- after all, black and white are the same whether you have 256 or 4096 shades of luminosity.

I also think that your comparison of 12-bit RAW vs. 8-bit JPG is informative and you draw a reasonable conclusion from the results. Although you did not allude to any comparison of 12 bits vs. 8 bits for non-lossy files, some readers might possibly make that assumption. However, I don't think that we should generalize the results to compare data recovery in a 12 bit file vs. a non-lossy file with 8 bit depth. The JPG file has discarded much of its data and, thus any sort of further processing involving shifting levels only leads to a lower quality image. I think that if a 12-bit RAW file were converted directly to its closely related TIFF file and then truncating the four LSB's that we would find recovering highlights is not too much worse than the results from a RAW file. However, the shadow areas would crash and burn because of the lack of detail as stated earlier by davesrose.

bill boehme wrote in post #3660732
I have been following this thread and have found it to be very interesting and informative. I certainly agree with your statement in the last paragraph that increasing the number of bits does not increase the dymanic range -- it just improves the fidelity of the shades of luminosity between black and white -- after all, black and white are the same whether you have 256 or 4096 shades of luminosity.

I also think that your comparison of 12-bit RAW vs. 8-bit JPG is informative and you draw a reasonable conclusion from the results. Although you did not allude to any comparison of 12 bits vs. 8 bits for non-lossy files, some readers might possibly make that assumption. However, I don't think that we should generalize the results to compare data recovery in a 12 bit file vs. a non-lossy file with 8 bit depth. The JPG file has discarded much of its data and, thus any sort of further processing involving shifting levels only leads to a lower quality image. I think that if a 12-bit RAW file were converted directly to its closely related TIFF file and then truncating the four LSB's that we would find recovering highlights is not too much worse than the results from a RAW file. However, the shadow areas would crash and burn because of the lack of detail as stated earlier by davesrose.

Yes, I failed to mention that the RAW file is a 12-bit-per-color file whereas the JPG is 8. As to how much data the JPG has "thrown away", I'm not sure. It is a "lossy" format, though the amount of discarded data varies on the degree of compression applied (adjustable by the "quality level"). But even if it were as non-lossy as RAW, it is still an 8-bit image as compared to RAW's 12-bit image, and would have to have less capability in some fashion. Whether that decreased capability was present in the form of less-smooth gradation in the dark areas of the image or in the "chopping off" of headroom, a compromise would have to be made.

An 8-stop range on an 8-bit image requires that data be divided something like this:

brightest stop - 128 levels of brightness per color (RGB)
stop 2 - 64 levels
stop 3 - 32 levels
stop 4 - 16 levels
stop 5 - 8 levels
stop 6 - 4 levels
stop 7 - 2 levels
darkest stop 8 - 1 levels

Now in real life, I believe that there is something of an "S" shape to the exposure curve, and that "may" translate into some variation towards the ends (possibly fewer levels at the top, and more at the darkest end) but I don't know this to be true. It's early in the AM and my mind is still foggy. And, I haven't really studied this in that great of depth.

If the full extent of a 12-bit RAW file is being used, it would divide something like this:

Brightest stop - 2048 levels
stop 2 - 1024 levels
stop 3 - 512 levels
stop 4 - 256 levels
stop 5 - 128 levels
stop 6 - 64 levels
stop 7 - 32 levels
stop 8 - 16 levels
stop 9 - 8 levels
stop 10 - 4 levels
stop 11 - 2 levels
darkest stop 12 - 1 level

Again, this is theoretical, and in practice, things could well be somewhat different. However, it is quite apparent that the 12-bit image has considerably more capability in terms of useable DR. If that full capability is used or not is another issue, but the "room" is there.

Anyway, in comparison to the 8-bit file, there's certainly the capability to present a greater DR without the dark regions becoming pixelated due to a lack of gradation. My tests indicate that for the 8-bit JPG file to maintain some degree of gradation in those darker stops, it gives up some capability at the brightest end of the spectrum.

I guess that the big question might be, 'how much of the highlight-recovery capability is given up due to the needs of going from a 12-bit to an 8-bit image and how much is given up due to losses inherent in the JPG format?'

davesrose wrote in post #3660197
Tonal range of an image can only be manipulated within the dynamic range of the image file itself. In a classical RGB color space, that's 256 shades of tone. Jpeg has a total of 256 luminosity levels to distribute....12bpc RAW has a total of 4092. Tom's examples are proof that jpeg has more limited tonal range then RAW....and here's a bit more:

Here are the terms for the different DRs of photography:

http://en.wikipedia.or​g/wiki/Exposure_range

I just found this article which describes the differences in contrast range and what that means in F stops between the different bit depths:

http://www.cambridgein​colour.com/tutorials/d​ynamic-range.htm

I asked you to prove that jpg has little tonal range past 4 stops, as you stated and I quoted, not to send me some links to wikipedia.

bill boehme wrote in post #3660732
I think that if a 12-bit RAW file were converted directly to its closely related TIFF file and then truncating the four LSB's that we would find recovering highlights is not too much worse than the results from a RAW file. However, the shadow areas would crash and burn because of the lack of detail as stated earlier by davesrose.

Well said, you clearly have a good understanding of the subject.

kpt4321 wrote in post #3661372
I asked you to prove that jpg has little tonal range past 4 stops, as you stated and I quoted, not to send me some links to wikipedia.

well apart from the obvious in the table that Tom showed (do you see that you only have 15 shades of tone past stop 4?),

OK, since you don't like to read....if you followed my link on dynamic range, it cites that with 8 bit A/D you get 8 stops of dynamic range (that's the dynamic range at the processor stage). 12bpc has 12 stops. See the difference? It has a disclaimer that that's the converter and not the image: where you would get less range because of noise. So we at least know that Jpegs are less then 8 stops Maybe I generalized too much when I said jpeg handles "around 4 stops"....but my point was it has less stops worth of tone then RAW. Since you seem to only want to flame me and not bother to read:

As an example, 10-bits of tonal precision translates into a possible brightness range of 0-1023 (since 210 = 1024 levels). Assuming that each A/D converter number is proportional to actual image brightness (meaning twice the pixel value represents twice the brightness), 10-bits of precision can only encode a contrast ratio of 1024:1. Most digital cameras use a 10 to 14-bit A/D converter, and so their theoretical maximum dynamic range is 10-14 stops. However, this high bit depth only helps minimize image posterization since total dynamic range is usually limited by noise levels. Similar to how a high bit depth image does not necessarily mean that image contains more colors, if a digital camera has a high precision A/D converter it does not necessarily mean it can record a greater dynamic range. In practice, the dynamic range of a digital camera does not even approach the A/D converter's theoretical maximum; 5-9 stops is generally all one can expect from the camera.

See the last line....I'll write it out for you: "In practice, the dynamic range of a digital camera does not even approach the A/D converter's theoretical maximum; 5-9 stops is generally all one can expect from the camera." That obviously must mean that's the final output data for RAW. In a jpeg image, you have 256 levels of contrast. That means it might not be able to logarithmically fill in stops 7, 8, or 9 with the data its having to process. Even in the "ideal" no noise world, jpeg is limited up to 8 stops. 12bpc color space gets up to 12 stops.

Tom W wrote in post #3660969
Now in real life, I believe that there is something of an "S" shape to the exposure curve, and that "may" translate into some variation towards the ends (possibly fewer levels at the top, and more at the darkest end) but I don't know this to be true. It's early in the AM and my mind is still foggy. And, I haven't really studied this in that great of depth.

Well we do know that in the ideal A/D, RAW gets its most amount of tone levels in the brightest areas. Most of your detail is in middle gray: which jpeg has 16 tones, and RAW has 64. I think it could be the logarithmic function of exposure: it's a beast that you get noise in your dark areas and less tones in your important contrast areas (middle gray). In your experiment, it's the detail of the map that really losses it in jpeg.....to me, that means the extra middle values of RAW are the most key.

Tom W wrote in post #3660969
I guess that the big question might be, 'how much of the highlight-recovery capability is given up due to the needs of going from a 12-bit to an 8-bit image and how much is given up due to losses inherent in the JPG format?'

The more I read up on this, the more I think higher bit depth helps reduce noise in the lower stops of "usable" range. We do know that highlights get the most values, and that the image's DR can be limited to the A/D bit depth (so say a 12bit camera converting down to 8 bit jpeg) as well as the sensor's light sensitivity range.

When you're adjusting levels on a jpeg, you're adjusting its tonal range: which has a smaller dynamic range then RAW's due to the processed bit depth.

I think what has confused everyone (myself included) is that we tend to think about "dynamic range" as being only a function of the camera sensor. There is a dynamic range with the total luminosity range of the scene you are photographing, another one for the light sensitivity of the sensor recording, and another one for the bit depth of the stored image (with jpeg having less range then the sensor is capable of and RAW having more range).