Dynamic Range

[Steen Bay]

It is simple. The dynamic range that is typically quoted in this forum is max sat signal divide by read noise.

For good reason!

But suppose you had a read noise of zero? Infinite dynamic range? No.

Well undefined actually, if you really achieve zero read noise.

A useful measure of photographic dynamic range is the max SNR to the lowest usable SNR.

If you have zero read noise then the max SNR is still undefined - back to the drawing book! ;-)

--
Its RKM

Sorry friend the min SNR is limited by the shot noise and there is never zero light when taking a photogarph. The whole point is the way DR is being defined here is not helpful The photographs IQ will be very much dependent on the min SNR that is present.

On DxO the 'noise floor' when calculating DR isn't the read noise alone, but SNR 1:1, so the shot noise is also taken into account (one photon is the lowest signal if the read noise is zero).

 

[Its RKM]

It is simple. The dynamic range that is typically quoted in this forum is max sat signal divide by read noise.

For good reason!

But suppose you had a read noise of zero? Infinite dynamic range? No.

Well undefined actually, if you really achieve zero read noise.

A useful measure of photographic dynamic range is the max SNR to the lowest usable SNR.

If you have zero read noise then the max SNR is still undefined - back to the drawing book! ;-)

--
Its RKM

Sorry friend the min SNR is limited by the shot noise and there is never zero light when taking a photogarph.

I didn't mention the min SNR, I referred to the max SNR .

The min SNR may well be limited by shot noise - the shot noise on that minumum level of light, which may well be less than the read noise. But there's no point in resorting to practicalities when your initial argument was an entirely hypothetical zero read noise.

--
Its RKM

 

[bigpigbig]

One more thing.

Most people speak of the highlight clipping as the main DR issue. I think it is equally important at both ends. Our eyes can more easily forgive clipping in the shadows when viewing a print as we can accept that the darkest shadows can be ink black while we can not accept that very very bright is white. What color is the sun? Not white. Snow is about the only thing I can think of that smack-you-in-the-eyes white.

But when I dodge dark areas of a negative I get a much more pleasing image than lifting -2.5 stop and darker areas digitally. Most of the time.

 

[Bernie Ess]

I'm satified with the dynamic range of my 5D, but if some company could somehow manage to create a smooth highlight roll-off similar to negative film, I would be ecstatic, even if the effective ISO was only 1600 (and I'd love to see ISO 25 included)

Fujis quite old S3pro and S5pro had exactly this: Two 6MP sets of pixels, one for "normal", one only for specular highlights. You could select the amount of blending in of the second highlight set of pixels. This gave those 3 DSLRs (which are outdated by todays body tech and features, and speed) an incredible latitude in the bright mids and lights (where it matters), whereas todays DSLRs high DR numbers are mainly in the shadows that are very clean, and DR goes down once the ISO is raised above base ISO.

Add the very low colour noise of the Fuji CCD and you got some extremely malleable files with a very analogue look that looks less "spectacular" than most modern DSLRs, but more natural and analogue.

Both have a Nikon mount and are available cheap on the second hand market.

Fuji's new and much more successful cameras (x100 and the new x-pro-1 have little in common with those older sensors, they are normal cmos sensors with slightly changed filters, DR in the lights is not even close to the older S3/S5)

Bernie

--

'All the technique in the world doesn’t compensate for the inability to notice.' (Elliot Erwitt)

 

[PhotoHawk]

Um - not quite. The K5 bests it by a bit.

 

[Great Bustard]

Um - not quite. The K5 bests it by a bit.

Quite:

http://www.dxomark.com/index.php/Cameras/Compare-Camera-Sensors/Compare-cameras-side-by-side/ (appareil1) 676%7C0 (brand) Pentax (appareil2) 680%7C0 (brand2) Nikon

Normalized to 8 MP:

K5: 14.1 stops at ISO 80, 13.7 stops at ISO 100
D7000: 13.9 stops at ISO 100

Per Pixel:

K5: 13.6 stops at ISO 80, 13.2 stops at ISO 100
D7000: 13.4 stops at ISO 100

They do, afterall, use the same sensor, so one wouldn't expect much difference.

 

[John Sheehy]

13 stops would be GREAT! There is really only about 7 stops of useable DR at present. Forget about the test numbers. They are not indicating usable tones.

Compared to what ? Did you look at film images 5 feet wide from 18 inches away?

As far as recovering shadows in post, sure there is detail there but even with the masterful 5D MKII if is very noisy detail below -2.5 stops.

The 5D2 is one of the worst cameras on the market for pulling up low-ISO shadows.

--
John

 

[John Sheehy]

On DxO the 'noise floor' when calculating DR isn't the read noise alone, but SNR 1:1, so the shot noise is also taken into account (one photon is the lowest signal if the read noise is zero).

Huh?

If we average 4 photons per pixel at 1/100, then we average:

2 photons at 1/200

1 photon at 1/400

1 photon every 2 pixels at 1/800

1 photon every 4 pixels at 1/1600

There is nothing special about 1 photon per pixel, whatsoever, except your interest with it.

There is no bottom to photon-only signals, except black or no light. Of course, if the exposure gets very low, the sensor may not catch any photons in some frames, but one or more in another.

--
John

 

[filmluvr]

Fujis quite old S3pro and S5pro had exactly this: Two 6MP sets of pixels, one for "normal", one only for specular highlights.

When I switched from film three years ago, I really wanted the Fuji body. However, the deal was nixed by the necessity of using Nikon lenses. Such a shame they never partnered with Canon...

--
http://jackandkelly.zenfolio.com/

 

[Great Bustard]

On DxO the 'noise floor' when calculating DR isn't the read noise alone, but SNR 1:1, so the shot noise is also taken into account (one photon is the lowest signal if the read noise is zero).

Huh?

If we average 4 photons per pixel at 1/100, then we average:

2 photons at 1/200

1 photon at 1/400

1 photon every 2 pixels at 1/800

1 photon every 4 pixels at 1/1600

There is nothing special about 1 photon per pixel, whatsoever, except your interest with it.

There is no bottom to photon-only signals, except black or no light. Of course, if the exposure gets very low, the sensor may not catch any photons in some frames, but one or more in another.

What Steen is saying is that since Photon Noise is everpresent and unavoidable, it doesn't make sense, in terms of the visual properties of the photo, to talk about a noise floor lower than 1 electron, since this represents a 100% NSR.

In fact, we could take it further and say that a 50% NSR or 25% NSR represents a more "meaninful" noise floor -- again, in terms of the visual properties of the final photo.

For example, let's consider a sensor with a per-pixel saturation of 4096 electrons. If the read noise is 1 electron, we have a DR of 12 stops. For a sensor with a read noise of 0.25 electrons, the DR is 14 stops. For a perfect sensor with no read noise, the DR is infinite.

If we take a photo, and push the shadows way, way, way up, are we going to see any meaningful differences between the three sensors? I mean, even the extreme shadows will be dominated by photon noise for read noise below one electron.

Thus, it is my position that using the read noise as the noise floor is not "useful", whereas using the 100% NSR is.

 

[Onabeach]

I agree DR is sadly lacking in the current crop of cameras.

When I'm on the beach, my eyes can easily see details in the white waves and the shadows AT THE SAME TIME! My camera has to blow the ocean in order to get any detail on the beach.

I suspect it is, like any thing else, a compromise. In order to get good live view/video performance, DR suffers. Take your pick.

 

[bigpigbig]

John,

What did you eat for breakfast?

At least the tone of your writing comes across as quite confrontational. Why? I am in it for the love of taking photos. There is no competition. Art for art's sake. If I misread you, my apologies.

13 stops would be GREAT! There is really only about 7 stops of useable DR at present. Forget about the test numbers. They are not indicating usable tones.

Compared to what ? Did you look at film images 5 feet wide from 18 inches away?

What does the viewing proximity have to do with DR? This is a serious question. I am not challenging you.

As far as recovering shadows in post, sure there is detail there but even with the masterful 5D MKII if is very noisy detail below -2.5 stops.

The 5D2 is one of the worst cameras on the market for pulling up low-ISO shadows.

Do you have a resource for this statement? "Worst camera on the market" Seriously?!

--
John

 

[kb2zuz]

The best prints are just under stops... most are 7-8. The best displays are around that too. So if you don't do some kind of post processing to blend in those areas (burn down the sky or such) you're going have to lower the contrast to fit 20 stops into an 8 stop print or photo. With B&W film, even if you had a perfect 14 stop negative (most were actually closer to 12 as I recall) you still had to squeeze those 14 stops into 8 if you printed it. Your options were:

• Print on low contrast paper (or with a low contrast filter for VC paper) and have a low contrast image (or as my first photo professor would say "it looks like mud")
• Print on high contrast paper (or high contrast filter for VC) and blow out the highlights, block-up the shadows, or both.
• Dodge and burn, possibly even with various contrast filters using VC paper in different areas and spend several hours getting a good print.

Yes, with digital, you could run tone mapping like what an HDR program does, but the tone mapping is what makes HDR look like HDR. So the only advantage such a camera would have over HDR is not having to take 3 shots.

Yes I agree the technology can get there and eventually provide higher bit depth with more dynamic range... but don't think that it's a cure all. You're still going to have to blend the images in post production... just like they do now, just like they did in the film days.

--
~K

 

[kb2zuz]

Seems too simple. Must be a good reason why it hasn't been done yet.

Been done: http://www.dpreview.com/reviews/fujifilms5pro/ (and the S3 and others before it). Reason it wasn't done again was the benfit wasn't worth the costs (monetary, resolution, and some other issues)

Not only would it allow for huge DR but Ultra Low noise.

Not really, it was worse than Canon options for high ISO noise at the time. It also had moiré issues, and it's buffer took longer to clear (processing time) because the files were huge. In many situations, it was a great camera, in others it was not.

While we are dreaming: if you could choose shutter speed for each. A 1/60 shutter speed where one sensor array stops collecting data after 1/250.

They are just getting to that point... but it's a few years from hitting the market.

--
~K

 

[filmluvr]

So the only advantage such a camera would have over HDR is not having to take 3 shots.

You make it sound so insignificant!

--
http://jackandkelly.zenfolio.com/

 

[Steen Bay]

On DxO the 'noise floor' when calculating DR isn't the read noise alone, but SNR 1:1, so the shot noise is also taken into account (one photon is the lowest signal if the read noise is zero).

Huh?

If we average 4 photons per pixel at 1/100, then we average:

2 photons at 1/200

1 photon at 1/400

1 photon every 2 pixels at 1/800

1 photon every 4 pixels at 1/1600

There is nothing special about 1 photon per pixel, whatsoever, except your interest with it.

There is no bottom to photon-only signals, except black or no light. Of course, if the exposure gets very low, the sensor may not catch any photons in some frames, but one or more in another.

I'm not saying that there's anything special about 1 photon per pixel, I'm just saying that the noise floor used by DxO is SNR 1:1, and not the read noise alone. That's how they have chosen to define DR, and that means that the lowest 'meaningful' signal is 1 photon per pixel if the read noise is zero. And of course, if we downsample the image to for example 1/10 of the original MP count, then the signal only needs to be 0.1 photon for each of the original pixels (if the read noise is zero).

 

[ohyva]

Is to use it more in the high-lights to make those more color accurate when close (current) clipping. This is IMHO the pain-point in digital photography since sensors are linear and thus clip hard when saturation limit is hit. I like the HTP in my Canon dSLR and use it a lot, but I'd like to get option for HTP2 and even HTP3 i.e. 1 to 2 stops more compared to current HTP(1).

So I do not see I neccessarily have to do anything special if Canon makes the proper tone curves for me. Some sort of adaptive contrast would also be nice i.e. the camera looks how wide the tone area is in the captured image and adjust it to make a pleasing image. And íf I do not like the result I always have the possibility to go to Lightroom and make my own adjustments.

The best prints are just under stops... most are 7-8. The best displays are around that too. So if you don't do some kind of post processing to blend in those areas (burn down the sky or such) you're going have to lower the contrast to fit 20 stops into an 8 stop print or photo. With B&W film, even if you had a perfect 14 stop negative (most were actually closer to 12 as I recall) you still had to squeeze those 14 stops into 8 if you printed it. Your options were:

• Print on low contrast paper (or with a low contrast filter for VC paper) and have a low contrast image (or as my first photo professor would say "it looks like mud")
• Print on high contrast paper (or high contrast filter for VC) and blow out the highlights, block-up the shadows, or both.
• Dodge and burn, possibly even with various contrast filters using VC paper in different areas and spend several hours getting a good print.

Yes, with digital, you could run tone mapping like what an HDR program does, but the tone mapping is what makes HDR look like HDR. So the only advantage such a camera would have over HDR is not having to take 3 shots.

Yes I agree the technology can get there and eventually provide higher bit depth with more dynamic range... but don't think that it's a cure all. You're still going to have to blend the images in post production... just like they do now, just like they did in the film days.

--
~K

 

[Its RKM]

If we take a photo, and push the shadows way, way, way up, are we going to see any meaningful differences between the three sensors? I mean, even the extreme shadows will be dominated by photon noise for read noise below one electron.

So, lets say you push the shadows to a level of 64, with the single photon resulting in and extra 6 levels. Are you really saying that you can't see the difference between a photo which has randomly distributed half the pixels at 70 and the remainder at 64 (1 photon shot noise) from one which only has a random quarter of the pixels at 70 (0.5 photon shot noise) or one which has random 1% of pixels at 70 (0.1 photon shot noise)?

I know which I would prefer and which would look lowest noise to me.
--
Its RKM

 

[PhotoHawk]

I stand by my earlier message. I'm going by the graphs on DXO as well. Even given that your point about the D7000 being far and away from the pack is in error witness the K5 performance. As you say both machines use the same sensor so I expect, given the minor variations from batch to batch they are for most intents and purposes of similar performance.

 

[Steen Bay]

On DxO the 'noise floor' when calculating DR isn't the read noise alone, but SNR 1:1, so the shot noise is also taken into account (one photon is the lowest signal if the read noise is zero).

Huh?

If we average 4 photons per pixel at 1/100, then we average:

2 photons at 1/200

1 photon at 1/400

1 photon every 2 pixels at 1/800

1 photon every 4 pixels at 1/1600

There is nothing special about 1 photon per pixel, whatsoever, except your interest with it.

There is no bottom to photon-only signals, except black or no light. Of course, if the exposure gets very low, the sensor may not catch any photons in some frames, but one or more in another.

What Steen is saying is that since Photon Noise is everpresent and unavoidable, it doesn't make sense, in terms of the visual properties of the photo, to talk about a noise floor lower than 1 electron, since this represents a 100% NSR.

In fact, we could take it further and say that a 50% NSR or 25% NSR represents a more "meaninful" noise floor -- again, in terms of the visual properties of the final photo.

For example, let's consider a sensor with a per-pixel saturation of 4096 electrons. If the read noise is 1 electron, we have a DR of 12 stops. For a sensor with a read noise of 0.25 electrons, the DR is 14 stops. For a perfect sensor with no read noise, the DR is infinite.

If we take a photo, and push the shadows way, way, way up, are we going to see any meaningful differences between the three sensors? I mean, even the extreme shadows will be dominated by photon noise for read noise below one electron.

Thus, it is my position that using the read noise as the noise floor is not "useful", whereas using the 100% NSR is.

That makes good sense. A couple of related threads here :

http://forums.dpreview.com/forums/read.asp?forum=1021&message=40450301

http://forums.dpreview.com/forums/read.asp?forum=1021&message=40451296