What exactly is a stop improvement in ISO?

[photogaz]

OK, so people say the 5D3 has a 1 stop improvement in RAW over the Mark II. However, what exactly is a stop.

For example, ISO3200 is double that of ISO1600, so that's a stop.
But ISO 800 is double of 400 so that's also a stop.

One is 400 difference, one is 1600 difference in a stop, so which stop are we talking here?

 

[eeek]

From what I've read is it is 2 stops. The III has as much noise on 25600 as the II at 6400.

edit: I'm sorry, I misread over where you said RAW. The 2 stop improvement is on JPEG's.

 

[chito]

you said it yourself.. a stop is double.. in iso, shutter speed and aperture.. from 1/30 to 1/60 there is a stop, fom 1/30 to 1/120 there are two stops.. get it?

 

[KeithR]

get it?

I think you're missing the OP's point, and it's a good one.

6400 ISO that looks like 3200 ISO is a far more impressive achievement than 800 that looks like 400, and getting one doesn't necessarily mean you'll get the other.

It's not clear which one we'll get a manufacturer talks about a one stop improvement...

 

[Meh]

get it?

I think you're missing the OP's point, and it's a good one.

6400 ISO that looks like 3200 ISO is a far more impressive achievement than 800 that looks like 400, and getting one doesn't necessarily mean you'll get the other.

It's not clear which one we'll get a manufacturer talks about a one stop improvement...

Oh, is that right?

 

[Tijn]

OK, so people say the 5D3 has a 1 stop improvement in RAW over the Mark II. However, what exactly is a stop.

e.
A 'stop' is a doubling or halving of light throughput. The ISO number is simply a standardized value for film/sensor light sensitivity (which is called the film 'speed', which by the way has nothing to do with processing or focus speed). Doubling this ISO value means doubling the light sensitivity (and thereby effectively the light throughput). A 1600 ISO sensor value will be twice as light-sensitive as an 800 ISO value, and an 800 ISO sensor value is twice as sensitive as ISO 400, which in turn is twice as sensitive as ISO 200. That means that ISO 1600 is three stops more sensitive than ISO 200. 200 -> 400 (1 stop) -> 800 (2 stops) -> 1600 (3 stops). Three stops difference means it's 2*2*2 = 8x as light-sensitive. With the same aperture and exposure time, pictures can be taken with 8x less light at a three stop higher ISO value.

The same 'stops' are used for the light throughput of the aperture. An aperture value of f/2 lets in twice as much light as an aperture value of f/2.8. Another halving of light throughput in aperture is f/4, then f/5.6. That means that an f/2 aperture is three stops more light-sensitive (three stops "faster") than f/5.6. This again means that it is 8x as light-sensitive, which means that it can take photos at similar shutter speeds and ISO values with 8x less light.

The catch for higher ISO values is the introduction of "noise" (errors in the sensor's light recording, resulting in coloured or brighter pixels than they would be with a 'clean' exposure). Therefore an ISO value may not mean the same as usable ISO. For example, the 12800 ISO extension of the 60D crop camera is VERY noisy. It will not produce usable images for big or even medium-sized prints. At the same time, 12800 ISO with the 5D mk III will be very usable, with relatively low amounts of noise. The same ISO value here only means that the film is equally light-sensitive, meaning that exposure will be the same for the two cameras when at the same ISO, aperture and in the same setting. The amount of noise introduced is vastly greater for one than for the other though, so there's a big difference between the usability of these two cameras' equal ISO values.

In this case, a one stop of ISO improvement means effective (usable) ISO in RAW files. That means that the 5D mk III will do ISO 25600 with comparable success (noise amount) as one stop lower ISO (ISO 12800, half as light-sensitive) would do on the 5D mk II.

 

[tt]

If the 5DmkIII looks a certain amount of ISO grain at 3200 - If its a 1 stop improvement on the 5DMkII then it looks similar to 5dMKII
At 1600

1 assumption is then that the MkIII looks better at any given ISO than MkII.
Usually I'd imagine talking about 1 stop improvement is focusing on the higher ISO end.

Eg - Say there's a certain level of graininess you accept.
Eg 1600 ISO at certain light shutter aperture etc with a 5DmkII.
For the same scene and settings,
A 1 stop improvement would be the MkIII looking similarly grainy, but at 3200 ISO.
(and looking better at 1600 ISO).

 

[Mt Spokane Photography]

From what I've read is it is 2 stops. The III has as much noise on 25600 as the II at 6400.

edit: I'm sorry, I misread over where you said RAW. The 2 stop improvement is on JPEG's.

Actually, Chuck Westfall said the two stops are im the sensor, not in the improved processing. I'm a bit skeptical, so I downloaded and printed the full resolution high iso samples from DPR. It certainly looks like two stops better from the sensor, but I'll wait for RAW tests to confirm it. Even the ISO 102400 prints were sharp and clear at 8.5 X 11. I've never seen any that good at that ISO.

 

[Meh]

From what I've read is it is 2 stops. The III has as much noise on 25600 as the II at 6400.

edit: I'm sorry, I misread over where you said RAW. The 2 stop improvement is on JPEG's.

Actually, Chuck Westfall said the two stops are im the sensor, not in the improved processing. I'm a bit skeptical, so I downloaded and printed the full resolution high iso samples from DPR. It certainly looks like two stops better from the sensor, but I'll wait for RAW tests to confirm it. Even the ISO 102400 prints were sharp and clear at 8.5 X 11. I've never seen any that good at that ISO.

He did say that and if it's fully true then we have a winner. It does make sense, lot's of tech improvements in the past 4 years since the 5D2, photosites are probably much larger and the gapless microlenses get a lot more light into those larger photosites. The high ISO images do look very good so far. Sooo good. And boom goes the dynamite.

 

[Janco]

But what does it mean that it will be 2 stops of improvements in iso/noise only for jpg's? I thought in RAW should be the same improvement possible, especially if improvement is in the sensor itself not in in-camera NR as I've read in another thread.... I think I miss something here...?

 

[Z]

... Actually, Chuck Westfall said the two stops are im the sensor, not in the improved processing...

I believe this is the quote you are referring to:

"He emphasizes that the bulk of the higher ISO image quality improvement comes from enhancements to the image sensor itself, as opposed to heavier-duty noise reduction being applied once the picture has been converted from analog to digital form."

Note that Chuck Westfall states the 'bulk' of the improvement comes from the sensor itself, not that it is entirely a sensor-mediated improvement. The bulk is a purposely ambiguous amount.

 

[Z]

Also... it's been quoted that the 5D Mark III has around a 2 stop improvement in high ISO shooting versus the mark II, but that the 1D X will retain around a stop more than the improvement seen in the 5D III.

Separately they have stated the 1D X has around a 2 stop improvement over the 1D mark IV. So which is it for the 1D X, 2 stops better than the APS-H 1D mark IV, or 3 stops better than the full frame 5D Mark II?

Marketing. Marketing. Marketing.

 

[Mt Spokane Photography]

Also... it's been quoted that the 5D Mark III has around a 2 stop improvement in high ISO shooting versus the mark II, but that the 1D X will retain around a stop more than the improvement seen in the 5D III.

Separately they have stated the 1D X has around a 2 stop improvement over the 1D mark IV. So which is it for the 1D X, 2 stops better than the APS-H 1D mark IV, or 3 stops better than the full frame 5D Mark II?

Marketing. Marketing. Marketing.

Canon rates the 1D MK IV at ISO 12800. Two stops better is 51200. http://www.usa.canon.com/cusa/consumer/products/cameras/slr_cameras/eos_1d_mark_iv

Canon rates the 5D MK II at ISO 6400. Two stops better is ISO 25600, three stops better is ISO 51200.
http://www.usa.canon.com/cusa/consumer/products/cameras/slr_cameras/eos_5d_mark_ii#Specifications

I think is pretty simple math.

 

[Z]

Canon rates the 1D MK IV at ISO 12800. Two stops better is 51200. http://www.usa.canon.com/cusa/consumer/products/cameras/slr_cameras/eos_1d_mark_iv

Canon rates the 5D MK II at ISO 6400. Two stops better is ISO 25600, three stops better is ISO 51200.
http://www.usa.canon.com/cusa/consumer/products/cameras/slr_cameras/eos_5d_mark_ii#Specifications

I should have clarified that I was referring to quotes about noise performance at various ISO speeds, not just native ISO range. Specifically mentioned in this article:

http://bobatkins.com/photography/digital/canon_eos_5D_MkIII_preview.html

"The [5D Mark III] pixel pitch is slightly smaller than that of the 5D MkII because the pixel count has been increased from 21MP to 22.3MP. Despite that, Canon claim that JPEGs from the EOS 5D MkIII at ISO 3200 show similar noise levels to JPEGs from the old EOS 5D MkII at ISO 800, a 2 stop advantage. While no numbers were provided for RAW files, it was said that although they don't show a 2 stop advantage, RAW files from the 5D MkIII do show lower intrinsic noise than those from the 5D MkII. It was noted that the larger pixels of the EOS 1D X along with whatever other "magic" Canon have come up with give the EOS 1D X an even lower noise level, around 1 stop better the the 5D MkIII for JPEG files."

So my point was, can the 1D X JPEGs be 3 stops better in terms of noise than the 5D II JPEGs and also 2 stops better than the 1D Mark IV JPEGs? According to DXO, the 5D Mark II sensor is around 0.5 stops better in terms of noise performance at equivalent ISO to the 1D Mark IV? The conclusion is either:

a) The 1D Mark IV has 1 stop better noise performance in JPEG than the 5D Mark II - it is a year younger so I suppose it's possible. TDP review suggests this is only possible with a disgusting amount of NR and loss of detail though.

b) It's marketing.

I think is [sic] pretty simple math.

Was that comment necessary?

 

[Meh]

Also... it's been quoted that the 5D Mark III has around a 2 stop improvement in high ISO shooting versus the mark II, but that the 1D X will retain around a stop more than the improvement seen in the 5D III.

Separately they have stated the 1D X has around a 2 stop improvement over the 1D mark IV. So which is it for the 1D X, 2 stops better than the APS-H 1D mark IV, or 3 stops better than the full frame 5D Mark II?

Marketing. Marketing. Marketing.

Yes, you're right to note that there is marketing spin to all of it. However, in referring to Chuck Westfall's statements he noted that the claimed 2 stop improvement in the 1DX over the 1D4 was in the JPEG not in the sensor. In the case of the 5D3 he has commented that most of the 2-stop improvement is in the sensor.

We have to be cautious when they refer to improvements for in-camera jpegs because the improvements Canon has added to its jpeg processing have already been available in post-processing noise reduction algorithms in Lightroom, DxO, etc. The advanced NR techniques were just not built into the cameras because they didn't have enough processing speed/power to do it while maintaining the frame rate.

This all makes a certain amount of sense. The 1DX sensor is only 2 years newer than the 1D4 sensor which was already one of the least noisiest sensors available. The 5D3 sensor is 4 years newer than the 5D2 sensor which was designed prior to a number of sensor advancements that we've seen in the 1D4 as well as in the Sony sensors which have remarkably low readout noise.

 

[Meh]

But what does it mean that it will be 2 stops of improvements in iso/noise only for jpg's? I thought in RAW should be the same improvement possible, especially if improvement is in the sensor itself not in in-camera NR as I've read in another thread.... I think I miss something here...?

Canon's official statements in marketing material only say 2 stops better noise but did not attribute it to sensor or jpeg processing. Chuck Westfall stated in interviews that in the case of the 1DX over the 1D4 the 2-stop improvement was mostly in the in-camera jpeg. He has said about the 5D3 that there is almost a 2 stop improvement in the sensor over the 5D2. If his statement about the 5D3 is correct there should be an even greater improvement in in-camera jpegs over the 5D2 if they built in the best NR algorithms... but maybe they didn't since there is only a single Digic 5+ processor whereas the 1DX has dual? I don't imagine they didn't use the same jpeg conversion algorithms as the 1DX but who knows.

 

[Janco]

Yes, you're right to note that there is marketing spin to all of it. However, in referring to Chuck Westfall's statements he noted that the claimed 2 stop improvement in the 1DX over the 1D4 was in the JPEG not in the sensor. In the case of the 5D3 he has commented that most of the 2-stop improvement is in the sensor.

We have to be cautious when they refer to improvements for in-camera jpegs because the improvements Canon has added to its jpeg processing have already been available in post-processing noise reduction algorithms in Lightroom, DxO, etc. The advanced NR techniques were just not built into the cameras because they didn't have enough processing speed/power to do it while maintaining the frame rate.

This all makes a certain amount of sense. The 1DX sensor is only 2 years newer than the 1D4 sensor which was already one of the least noisiest sensors available. The 5D3 sensor is 4 years newer than the 5D2 sensor which was designed prior to a number of sensor advancements that we've seen in the 1D4 as well as in the Sony sensors which have remarkably low readout noise.

Thank you, that answered my question quite well. Didn't think of postprocessing software improvements compared to in-camera NR...

 

[Mt Spokane Photography]

Canon rates the 1D MK IV at ISO 12800. Two stops better is 51200. http://www.usa.canon.com/cusa/consumer/products/cameras/slr_cameras/eos_1d_mark_iv

Canon rates the 5D MK II at ISO 6400. Two stops better is ISO 25600, three stops better is ISO 51200.
http://www.usa.canon.com/cusa/consumer/products/cameras/slr_cameras/eos_5d_mark_ii#Specifications

I should have clarified that I was referring to quotes about noise performance at various ISO speeds, not just native ISO range. Specifically mentioned in this article:

http://bobatkins.com/photography/digital/canon_eos_5D_MkIII_preview.html

"The [5D Mark III] pixel pitch is slightly smaller than that of the 5D MkII because the pixel count has been increased from 21MP to 22.3MP. Despite that, Canon claim that JPEGs from the EOS 5D MkIII at ISO 3200 show similar noise levels to JPEGs from the old EOS 5D MkII at ISO 800, a 2 stop advantage. While no numbers were provided for RAW files, it was said that although they don't show a 2 stop advantage, RAW files from the 5D MkIII do show lower intrinsic noise than those from the 5D MkII. It was noted that the larger pixels of the EOS 1D X along with whatever other "magic" Canon have come up with give the EOS 1D X an even lower noise level, around 1 stop better the the 5D MkIII for JPEG files."

So my point was, can the 1D X JPEGs be 3 stops better in terms of noise than the 5D II JPEGs and also 2 stops better than the 1D Mark IV JPEGs? According to DXO, the 5D Mark II sensor is around 0.5 stops better in terms of noise performance at equivalent ISO to the 1D Mark IV? The conclusion is either:

a) The 1D Mark IV has 1 stop better noise performance in JPEG than the 5D Mark II - it is a year younger so I suppose it's possible. TDP review suggests this is only possible with a disgusting amount of NR and loss of detail though.

b) It's marketing.

I think is [sic] pretty simple math.

Was that comment necessary?

Sorry, if I offended you.

I'm not defending Canon, merely pointing out that their claim of HIGH ISO improvement is based on their previous claims, not those of someone else.

 

[Viggo]

This is some of the reason I don't like the Dxo tests. As it states that the 5d2 is better at noise then athe mk4. Now, if you have tried these cameras you know it's completely wrong. On one hand...

Between 100-800 the 5d kills the mk4... But from 1600 and up the mk4 demolishes the 5d2. So seeing that the 5d shows noise at 1320 something and the mk4 is 1260 or whatever is just to take it way out of contex.....

Very valid question from the OP. And there is different answers...

I think Canon states two stops as being able to shoot at twice and four times the iso. And they claim the same noise at the highest setting being better than two stops down on the 5d2, which could easily be true.

Now take Lightroom in to the equation, Lr3 and the 5d2 vs the 5d3 with Lr4 gives you easily three stops, so "only for jpeg" two stops is only valid with the DPP software...

Yes I am drawing my own conclusions here, and claiming stuff, because it's friday and I'm enjoying a few beers, cheers everybody! Have a good one!

 

[Fleetie]

Yes I am drawing my own conclusions here, and claiming stuff, because it's friday

Is it?!

Where on Earth are you?!