Interesting comparison of FF and medium format

[xpatUSA]

I understand (I think) that one needs more than 8 bits to encompass the gamut visible to people ...

I'm not sure that "encompass" is the right word, Dan, because it implies that a greater bit-depth gives a larger gamut.

However, on the 1931 CIE "horseshoe" diagram, the bit-depth certainly affects the number of steps available between, for example, the white point and a primary or indeed between any two visible colors.

[DanK]

Ted,

Thanks. Point taken. However, I still do wonder whether humans, withough machines to aid them, can distinguish the additional gradations one gets from going from 4 billion to 35 billion colors. Has anyone ever tested this?

Dan

[xpatUSA]

Ted,

Thanks. Point taken. However, I still do wonder whether humans, without machines to aid them, can distinguish the additional gradations one gets from going from 4 billion to 35 billion colors. Has anyone ever tested this?

Dan

I believe that there is a wealth of information and test results but not about the exact question raised by your good self above.

For example there is the "Just Noticeable Difference" (JND) in colors seen by humans as it relates to the various measurements of delta-E (color difference) over the years - often said to have a value of 1 in terms of L*a*b* scaling; sometimes just referring to the two hue axes and vectorial measure; sometimes including Lightness, therefore a 3-axis vector. I've also see 2.3 mentioned as a value for JND.

Point being that two colors at a four-billionth difference would be ridiculously similar and certainly way, way less than one JND and maybe that answers your question right there?

[DanK]

Indeed. And that’s only 14 bit


Sent from my iPad using Tapatalk

[dje]

And in reality you'll never actually view an image in 14 or 16 bit color. The average monitor is 8 bit with some high end models 10 bit. I can't speak with any authority about printing but I think similar constraints apply. Dithering helps with banding too.

One area where bit depth becomes more important is High Dynamic Range imaging, particularly where transmission bandwidth is important ie video. HDR monitors can typically produce luminance levels in the range 400-1000 cd/sqm compared to standard monitors which are about 100-200 cd/sqm max. With the stretched luminance range, banding is more of an issue and a bit depth of 10 or 12 is considered desirable. The JND's of interest are mainly with luminance rather than hue.

Where bandwidth is not important, HDR is best done with floating point data.

Dave

[DanK]

My question was simplistic. The issue is less whether one can see the difference between 14-bit and 16-bit images as captured than whether one can see a difference after processing. The answer to the first question is almost certainly "no", but the point of capturing additional bit depth is primarily to provide headroom in editing--in particular, to avoid banding when editing pulls the values of two colors apart.

This morning I found an excellent article about this by Greg Benz, https://petapixel.com/2018/09/19/8-1...u-really-need/. He compares 12-bit to 14-bit, not 14-bit to 16-bit, but his results are instructive anyway. He did find differences for extreme edits of deep shadows:

Ultimately, I find that at ISO 64 with a Nikon D850:

12-bit files can be pushed 3-4 stops in LR or CO
14-bit files can be pushed 5-6 stops in LR or 4-5 stops in CO

This implies that the difference between 14 and 16 bits might be visible, but only with truly extreme edits. I doubt the images in the Lensrentals blog were subjected to edits anywhere nearly that severe.

[Manfred M]

And in reality you'll never actually view an image in 14 or 16 bit color. The average monitor is 8 bit with some high end models 10 bit. I can't speak with any authority about printing but I think similar constraints apply. Dithering helps with banding too.

One area where bit depth becomes more important is High Dynamic Range imaging, particularly where transmission bandwidth is important ie video. HDR monitors can typically produce luminance levels in the range 400-1000 cd/sqm compared to standard monitors which are about 100-200 cd/sqm max. With the stretched luminance range, banding is more of an issue and a bit depth of 10 or 12 is considered desirable. The JND's of interest are mainly with luminance rather than hue.

Where bandwidth is not important, HDR is best done with floating point data.

Dave

This is why I use medium sized prints when looking at these images. The are generally done with wide gamut printers that exceed the AdobeRGB colour space, are generally upsampled and not downsampled, are independent of the setting and quality issues associated with computer screens, etc.

[xpatUSA]

Originally Posted by dje And in reality you'll never actually view an image in 14 or 16 bit color. The average monitor is 8 bit with some high end models 10 bit. I can't speak with any authority about printing but I think similar constraints apply. Dithering helps with banding too.

One area where bit depth becomes more important is High Dynamic Range imaging, particularly where transmission bandwidth is important ie video. HDR monitors can typically produce luminance levels in the range 400-1000 cd/sqm compared to standard monitors which are about 100-200 cd/sqm max. With the stretched luminance range, banding is more of an issue and a bit depth of 10 or 12 is considered desirable. The JND's of interest are mainly with luminance rather than hue.

This is why I use medium sized prints when looking at these images. The are generally done with wide gamut printers that exceed the AdobeRGB colour space, are generally upsampled and not downsampled, are independent of the setting and quality issues associated with computer screens, etc.

Assuming that Manfred is referring refer to HDR images, I am curious as to how gamut-width affects dynamic range; or vice-versa?

Where bandwidth is not important, HDR is best done with floating point data.

Yes, I always edit in 32-bit floating point ... not because I have to, but because RawTherapee's default is that. I also have it set thus in the GIMP.

[xpatUSA]

... but the point of capturing additional bit depth is primarily to provide headroom in editing--in particular, to avoid banding when editing pulls the values of two colors apart.

This morning I found an excellent article about this by Greg Benz, https://petapixel.com/2018/09/19/8-1...u-really-need/. He compares 12-bit to 14-bit, not 14-bit to 16-bit, but his results are instructive anyway ...

Good article which makes the point. Apart from:

"Which means that an 8-bit RGB image in Photoshop will have a total of 24-bits per pixel (8 for red, 8 for green, and 8 for blue). A 16-bit RGB or LAB image in Photoshop would have 48-bits per pixel, etc."

Dare I even suggest that adding bit-depths together like that is semantically incorrect - although common enough in the Lexicon of digital photography. I guess we all know what it means but it remains incorrect.

I could prove it - but I'm told that most photographers would find that revelation of limited use.

[xpatUSA]

Assuming that Manfred is referring refer to HDR images, I am curious as to how gamut-width affects dynamic range; or vice-versa?

Anybody?

[Manfred M]

Assuming that Manfred is referring refer to HDR images, I am curious as to how gamut-width affects dynamic range; or vice-versa?

Definitely not looking at HDR images. In my view, dynamic range is a bit of a red herring to begin with. While we have camera manufacturers out there claiming that they have a 15-stop dynamic range, some of that range is more marketing than useful.

Most better computer screens have a dynamic range in the 10 stop range (1000:1 contrast ratio is pretty typical). Prints have an even lower one; I am told that matte cotton rag papers are around 7.5 stops and glossier surfaces (like baryta papers) are in the 8-stop range. So all that dynamic range that sensors can capture can not be realized by our output devices.

That does not even get into issues with the human visual system; we are most sensitive to the mid-tones. The details in the deepest shadows and brightest highlights are hard to distinguish.

Lets also not forget that to get that high dynamic range on our screens, we need to be working in relatively dimly lit rooms. Ideally below 32 lux with screen white point set to between 80 and 120 cd/square meter (reference is from Tom P Ashe's book Color Management & Quality Output; Focal Press 2014).

[xpatUSA]

Definitely not looking at HDR images. In my view, dynamic range is a bit of a red herring to begin with. While we have camera manufacturers out there claiming that they have a 15-stop dynamic range, some of that range is more marketing than useful.

OK. It wasn't immediately obvious to me.

For what it's worth, my favorite camera does 9EV at best and that beats my monitor, which is dutifully set per your recommendations.

[DanK]

In my view, dynamic range is a bit of a red herring to begin with. While we have camera manufacturers out there claiming that they have a 15-stop dynamic range, some of that range is more marketing than useful.

Most better computer screens have a dynamic range in the 10 stop range (1000:1 contrast ratio is pretty typical). Prints have an even lower one; I am told that matte cotton rag papers are around 7.5 stops and glossier surfaces (like baryta papers) are in the 8-stop range. So all that dynamic range that sensors can capture can not be realized by our output devices.

I think this misses the point. The benefit of a greater dynamic range in a camera is not to increase the dynamic range of the output. It's to avoid truncation of the data.

Imagine a scene with a range of 13 EV captured by a camera with a DR of 11 EV. Two stops of data are lost, and the corresponding regions will be detail-free black or white. Redo the capture with a camera that has a range of 14 EV, and all of the scene will be represented in the output. It will be scaled to fit within the software's 0-255 tonal range.

I'm not sure where precisely where in the pipeline the rescaling occurs, but it is complete by the time the raw processing software renders the image.

[Manfred M]

I think this misses the point. The benefit of a greater dynamic range in a camera is not to increase the dynamic range of the output. It's to avoid truncation of the data.

Imagine a scene with a range of 13 EV captured by a camera with a DR of 11 EV. Two stops of data are lost, and the corresponding regions will be detail-free black or white. Redo the capture with a camera that has a range of 14 EV, and all of the scene will be represented in the output. It will be scaled to fit within the software's 0-255 tonal range.

I'm not sure where precisely where in the pipeline the rescaling occurs, but it is complete by the time the raw processing software renders the image.

Dan - to a certain extent I agree, but I also know that 15-stops of dynamic range is not generally needed when my final print is done on a medium where the dynamic range is around 8-stops. Techniques like ETTR will certainly benefit from more data at the light end of the exposure.

I rarely have had a problem with image quality that way unless I have really blown my exposure.

[DanK]

Manfred,

I think you are missing the point. What's relevant is the DR of the scene and the camera, not the DR of the output medium. ETTR does not help if the DR of the scene exceeds the DR of the camera. There is simply no way to avoid blowing the exposure if the DR of the scene exceeds the DR of the camera. ETTR will lessen truncation but not eliminate it. That's why people bracket exposures.

it's not terribly surprising that you haven't run into problems with this, given the types of photography you do and the fact that for some years, high-end Nikons have been among the best cameras in terms of DR. When I was shooting urban night photography with a 5DIII (blue line), I had to bracket more often than people using cameras like the D850 (black line). That's one reason why I upgraded to the otherwise very similar 5D IV. This has nothing to do with the output medium.

[Manfred M]

Manfred,

I think you are missing the point. What's relevant is the DR of the scene and the camera, not the DR of the output medium. ETTR does not help if the DR of the scene exceeds the DR of the camera. There is simply no way to avoid blowing the exposure if the DR of the scene exceeds the DR of the camera. ETTR will lessen truncation but not eliminate it. That's why people bracket exposures.

it's not terribly surprising that you haven't run into problems with this, given the types of photography you do and the fact that for some years, high-end Nikons have been among the best cameras in terms of DR. When I was shooting urban night photography with a 5DIII (blue line), I had to bracket more often than people using cameras like the D850 (black line). That's one reason why I upgraded to the otherwise very similar 5D IV. This has nothing to do with the output medium.

Understood Dan and I have shot a lot of images with my Panasonic GX-7 that did not have great dynamic range or low noise performance. I tend to try to shoot with as low an ISO setting as I can get away with as that does maximize dynamic range. I also try to shoot in lighting where dynamic range is not an issue (I rarely point my camera directly toward the sun). I also bracket when I shoot in challenging situations as I can usually build a composite image where I get the details I am looking for, even when hand holding.

[lunaticitizen]

Manfred,

I think you are missing the point. What's relevant is the DR of the scene and the camera, not the DR of the output medium. ETTR does not help if the DR of the scene exceeds the DR of the camera. There is simply no way to avoid blowing the exposure if the DR of the scene exceeds the DR of the camera. ETTR will lessen truncation but not eliminate it. That's why people bracket exposures.

it's not terribly surprising that you haven't run into problems with this, given the types of photography you do and the fact that for some years, high-end Nikons have been among the best cameras in terms of DR. When I was shooting urban night photography with a 5DIII (blue line), I had to bracket more often than people using cameras like the D850 (black line). That's one reason why I upgraded to the otherwise very similar 5D IV. This has nothing to do with the output medium.

Not that this contradicts anything you wrote, but the "DR" you mentioned is what Mr. William J. Claff of photonstophotos.net calls Photographic Dynamic Range (PDR), not the classical definition of dynamic range.

The measurement of the usual dynamic range -which what Mr. Claff calls Engineering Dynamic Range (EDR)- can be viewed from the Input-referred Read Noise versus ISO Setting page. You'll have to select the camera and click on its legend on the right side of the chart, though.

[xpatUSA]

Not that this contradicts anything Dan wrote, but the "DR" he mentioned is what Mr. William J. Claff of photonstophotos.net calls Photographic Dynamic Range (PDR), not the classical definition of dynamic range.

Which would have been https://www.iso.org/standard/59420.html., but it is withdrawn.

Doug Kerr had plenty to say about the 2003 version:

http://dougkerr.net/Pumpkin/articles...amic_range.pdf

So what is the classical definition? Is is still log2(saturation over noise=mean signal)?

[lunaticitizen]

Yes, and it's the only definition I know. Some people say that the choice of SNR=1 is arbitrary. I noticed that Doug Kerr also mentioned the same thing.

Jack Hogan also wrote about the definition of DR here.
https://www.strollswithmydog.com/eng...e-photography/

Bill Claff also wrote a primer about his PDR and the usual DR here.
https://photonstophotos.net/GeneralT...amic_Range.htm

I didn't understand the short primer when I read it for the first time, though. I got to understand PDR by reading this book.
https://iopscience.iop.org/book/978-0-7503-2558-5

[xpatUSA]

Yes, and it's the only definition I know. Some people say that the choice of SNR=1 is arbitrary. I noticed that Doug Kerr also mentioned the same thing.

Leo,

I would agree about 'arbitrary' especially because for a normal distribution at SNR equals 1, it has negative values of signal in it, as Kerr said. I find the trick ISO uses to project the measured noise value at 1 percent signal downward a bit of a crock, if you know what I mean.

And, if you've ever seen an image with areas where SNR equals 1, you yourself might not consider those areas acceptable. I don't. I've seen mentions of much higher SNRs being acceptable - for example, 5 or 10 for recording artwork.

Jack Hogan also wrote about the definition of DR here.
https://www.strollswithmydog.com/eng...e-photography/

Bill Claff also wrote a primer about his PDR and the usual DR here.
https://photonstophotos.net/GeneralT...amic_Range.htm

I didn't understand the short primer when I read it for the first time, though. I got to understand PDR by reading this book.
https://iopscience.iop.org/book/978-0-7503-2558-5

Thanks for the link. Yes, they certainly know their stuff. I have corresponded with both of them several times. Kasson too:

https://blog.kasson.com/
.