How much resolution decrease as we raise the ISO ?

Started May 18, 2012 | Discussions
Silverback1988
Contributing MemberPosts: 689
Like?
How much resolution decrease as we raise the ISO ?
May 18, 2012

Hi,
I curious if someone of you has been calculate this.

As we know, if we keep the noise constant by applying NR, then there would be a decrease in resolution as we raise the ISO.
But by how much from the native resolution ?

For example we had 20 MP camera.
Say we are able to get the full resolution of 20 MP is on ISO 100.

As we raise the ISO, say ISO 1600, then we keep the noise similar to the ISO 100 by applying NR. How much resolution is left ? Surely is not 20 MP anymore.
Would be it 5 MP only ?

I think we could compare it with a 5 MP camera and then shot with ISO 100.
If the picture is identical then the resolution is the same.

And I think the decrease is not linear between ISO 100 and ISO 400.

Opinions are welcome.

Sorry, English is my third language
I hope I state my question clearly.

Thanks for the reply.

-
Brian

PBear2
Junior MemberPosts: 37
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to Silverback1988, May 18, 2012

Check out Philip Bloom's web page, he states the most usable non noisy isos are 160, 320,640,1250, 2500,

the most noisy are the mulyiples of 125, 250, 500, 1000

Reply   Reply with quote   Complain
Rick Knepper
Veteran MemberPosts: 9,598Gear list
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to PBear2, May 18, 2012

Could you provide the name of the article or the link? Using the key words in your post, searches of his web site do not produce anything useful. As for native ISOs e.g. 100, 200, 400 etc what was his opinion?

PBear2 wrote:

Check out Philip Bloom's web page, he states the most usable non noisy isos are 160, 320,640,1250, 2500,

the most noisy are the mulyiples of 125, 250, 500, 1000

-- hide signature --

Rick Knepper, photographer, photography never for sale, check my profile for gear list and philosophy.

 Rick Knepper's gear list:Rick Knepper's gear list
Nikon D3X Nikon D800E Canon EOS 5D Mark III Canon EOS 6D Canon TS-E 17mm f/4L +16 more
Reply   Reply with quote   Complain
asad137
Contributing MemberPosts: 686
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to Rick Knepper, May 18, 2012

Rick Knepper wrote:

Could you provide the name of the article or the link? Using the key words in your post, searches of his web site do not produce anything useful. As for native ISOs e.g. 100, 200, 400 etc what was his opinion?

I found it quite easily searching for "Philip Bloom iso":

http://wiegaertnerfilms.com/tutorials/the-best-iso-settings-for-canon-video-dslrs/

This is a video test, but I would assume the results would translate to stills.

Reply   Reply with quote   Complain
asad137
Contributing MemberPosts: 686
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to Rick Knepper, May 18, 2012

Oops, this is the Philip Bloom link (first result when searching for "philip bloom iso":

http://philipbloom.net/2010/12/16/shooting-video-with-a-dslr/

The other link has a nice video showing the noise differences between ISOs.

Reply   Reply with quote   Complain
DSPographer
Senior MemberPosts: 2,232
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to Silverback1988, May 18, 2012

Silverback1988 wrote:

Hi,
I curious if someone of you has been calculate this.

As we know, if we keep the noise constant by applying NR, then there would be a decrease in resolution as we raise the ISO.
But by how much from the native resolution ?

First consider how the noise in a pixel varies with exposure and pixel size. Emil explains that here:
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html

So, we see that shot noise depends on the amount of light. So, when read noise is negligible, reducing resolution by a factor of 2- or reducing pixel count by a factor of 4- has the same effect on signal to noise as increasing exposure by a factor of 4.

But, noise reduction algorithms are usually non-linear. They keep fine details above a certain contrast and suppress them below that contrast. Then increasing ISO doesn't change the resolution for all contrast: instead it changes the threshold of how strong the contrast needs to be to be above the noise, and so make it past the noise suppression algorithm. So, for shadow details with such slight contrast that they are just above the noise at a given exposure, the detail in megapixels changes as the inverse of the exposure ISO setting increase. But for strong contrast details in bright areas, the resolution doesn't go down until the exposure reduction causes the noise level to approach the contrast of the details.

Reply   Reply with quote   Complain
Pultzar
Senior MemberPosts: 1,424
Like?
Clipped highlights
In reply to PBear2, May 18, 2012

ISO 160 is really ISO 200 and then a 1/3rd stop pull. So you may have less noise at the expense of clipping highlights.

Reply   Reply with quote   Complain
bronxbombers
Forum ProPosts: 18,226
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to Silverback1988, May 18, 2012

Silverback1988 wrote:

Hi,
I curious if someone of you has been calculate this.

As we know, if we keep the noise constant by applying NR, then there would be a decrease in resolution as we raise the ISO.
But by how much from the native resolution ?

For example we had 20 MP camera.
Say we are able to get the full resolution of 20 MP is on ISO 100.

As we raise the ISO, say ISO 1600, then we keep the noise similar to the ISO 100 by applying NR. How much resolution is left ? Surely is not 20 MP anymore.
Would be it 5 MP only ?

I think we could compare it with a 5 MP camera and then shot with ISO 100.
If the picture is identical then the resolution is the same.

And I think the decrease is not linear between ISO 100 and ISO 400.

Opinions are welcome.

Sorry, English is my third language
I hope I state my question clearly.

Thanks for the reply.

-
Brian

Hard to say, it might depends upon the scene too. What if important parts in shadow vs in brighter areas, etc. Even at a given ISO different regions of the photo may have different details levels, look in ISO100 shadows on 5D3 and not so much fine detail.

But assuming we just talk about middle gray yeah sorry I don't know. Not sure anyone has bothered carrying it out either a straight compare or a NR comparison (the latter is tricky since NR tech constantly changes, differently people have different ideas on any given NR result and NR works different scene to scene).

My very unscientific, very rough feeling was that above ISO1600 my 7D didn't have, on average, all that much reach advantage of my 5D2 anymore (but I reserve the right to change my opinion based on a fair, direct comparison).

Reply   Reply with quote   Complain
bronxbombers
Forum ProPosts: 18,226
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to bronxbombers, May 18, 2012

bronxbombers wrote:

Silverback1988 wrote:

Hi,
I curious if someone of you has been calculate this.

As we know, if we keep the noise constant by applying NR, then there would be a decrease in resolution as we raise the ISO.
But by how much from the native resolution ?

For example we had 20 MP camera.
Say we are able to get the full resolution of 20 MP is on ISO 100.

As we raise the ISO, say ISO 1600, then we keep the noise similar to the ISO 100 by applying NR. How much resolution is left ? Surely is not 20 MP anymore.
Would be it 5 MP only ?

I think we could compare it with a 5 MP camera and then shot with ISO 100.
If the picture is identical then the resolution is the same.

And I think the decrease is not linear between ISO 100 and ISO 400.

Opinions are welcome.

Sorry, English is my third language
I hope I state my question clearly.

Thanks for the reply.

-
Brian

Hard to say, it might depends upon the scene too. What if important parts in shadow vs in brighter areas, etc. Even at a given ISO different regions of the photo may have different details levels, look in ISO100 shadows on 5D3 and not so much fine detail.

But assuming we just talk about middle gray yeah sorry I don't know. Not sure anyone has bothered carrying it out either a straight compare or a NR comparison (the latter is tricky since NR tech constantly changes, differently people have different ideas on any given NR result and NR works different scene to scene).

My very unscientific, very rough feeling was that above ISO1600 my 7D didn't have, on average, all that much reach advantage of my 5D2 anymore (but I reserve the right to change my opinion based on a fair, direct comparison).

(and i bet it is probably more like ISO3200+)

Reply   Reply with quote   Complain
bronxbombers
Forum ProPosts: 18,226
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to bronxbombers, May 18, 2012

actually for straight up compare it might be just like 4 stops for 1/2 the detail

bronxbombers wrote:

Silverback1988 wrote:

Hi,
I curious if someone of you has been calculate this.

As we know, if we keep the noise constant by applying NR, then there would be a decrease in resolution as we raise the ISO.
But by how much from the native resolution ?

For example we had 20 MP camera.
Say we are able to get the full resolution of 20 MP is on ISO 100.

As we raise the ISO, say ISO 1600, then we keep the noise similar to the ISO 100 by applying NR. How much resolution is left ? Surely is not 20 MP anymore.
Would be it 5 MP only ?

I think we could compare it with a 5 MP camera and then shot with ISO 100.
If the picture is identical then the resolution is the same.

And I think the decrease is not linear between ISO 100 and ISO 400.

Opinions are welcome.

Sorry, English is my third language
I hope I state my question clearly.

Thanks for the reply.

-
Brian

Hard to say, it might depends upon the scene too. What if important parts in shadow vs in brighter areas, etc. Even at a given ISO different regions of the photo may have different details levels, look in ISO100 shadows on 5D3 and not so much fine detail.

But assuming we just talk about middle gray yeah sorry I don't know. Not sure anyone has bothered carrying it out either a straight compare or a NR comparison (the latter is tricky since NR tech constantly changes, differently people have different ideas on any given NR result and NR works different scene to scene).

My very unscientific, very rough feeling was that above ISO1600 my 7D didn't have, on average, all that much reach advantage of my 5D2 anymore (but I reserve the right to change my opinion based on a fair, direct comparison).

Reply   Reply with quote   Complain
bronxbombers
Forum ProPosts: 18,226
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to asad137, May 18, 2012

asad137 wrote:

Rick Knepper wrote:

Could you provide the name of the article or the link? Using the key words in your post, searches of his web site do not produce anything useful. As for native ISOs e.g. 100, 200, 400 etc what was his opinion?

I found it quite easily searching for "Philip Bloom iso":

http://wiegaertnerfilms.com/tutorials/the-best-iso-settings-for-canon-video-dslrs/

This is a video test, but I would assume the results would translate to stills.

video has tone curves and all baked in so the result don't appy to raw stills, the iso160 and 200 are the same thing, if you apply -1/3 ec to iso160 you get same as if you used iso200 or if you used iso200 and applied ec +1/3 you get exact same final result as if you shot iso160 (the raw range is a bit different but the final result is exactly the same, they measure to the same dynmaic range and middle gray SNR)

Reply   Reply with quote   Complain
PBear2
Junior MemberPosts: 37
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to Rick Knepper, May 19, 2012

sorry, it was under the Education tab section, initial set ups for DSLRS, he suggested manual iso settings rather than Auto iso, and then set up a USER defined Picture Style: Neutral, sharpness off, contrast off, saturation -2 then work from there

The blacks in my stills changed, the video looked better, not scientific but I like how it looks.

Reply   Reply with quote   Complain
Silverback1988
Contributing MemberPosts: 689
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to bronxbombers, May 19, 2012

Hi bronxbombers,

bronxbombers wrote:

Hard to say, it might depends upon the scene too. What if important parts in shadow vs in brighter areas, etc. Even at a given ISO different regions of the photo may have different details levels, look in ISO100 shadows on 5D3 and not so much fine detail.

Yeah the banding is there

But assuming we just talk about middle gray yeah sorry I don't know. Not sure anyone has bothered carrying it out either a straight compare or a NR comparison (the latter is tricky since NR tech constantly changes, differently people have different ideas on any given NR result and NR works different scene to scene).

Can we just make a certain amount of noise as a standard then calculate the resolution ?

For example, just take the noise level (at pixel level) of D700 or 5D on ISO 100 as a standard value.Then pick an image from other camera (or the same camera) but on other ISO (say 1600), applying some NR (or with some downsampling), so the noise level is the same as the standard one. Then we could calculate how much resolution is left in that ISO.
Of course with more advanced NR in the future, the resolution will be improved.
That is a good thing.

My very unscientific, very rough feeling was that above ISO1600 my 7D didn't have, on average, all that much reach advantage of my 5D2 anymore (but I reserve the right to change my opinion based on a fair, direct comparison).

So we could say that at ISO 1600, 18 MP from 7D has the same resolution as 8 MP (after cropping) from 5D2.

Another example (a roughly guess too)
If we take 5D2 as the standard ,
the ISO 100 from 5D2 is considered as 21 MP,
the ISO 200 from 5D2 is calculated as 20 MP,
the ISO 400 from 5D2 is calculated as 17 MP,
the ISO 1600 from 5D2 is calculated as 5 MP,
(the numbers above are just a guess, not the actual calculated resolution)

then
combined with your experience,
the ISO 1600 from 7D is only 2 MP worth. (5 MP divided by 2.56 = 1.95)

In other word, the 5D2 at ISO 100 (or even ISO 400) has about 3 times bigger (linear) print than 7D at ISO 1600.

If the file from 5D2 at ISO 100 being printed at 30 cm x 45 cm,
then the 7D at ISO 1600 will be as good but in 10 cm x 15 cm print.

That's quite close from my experience (not a scientific one of course)

Anyone done the actual comparison yet ?

-
Brian

Reply   Reply with quote   Complain
Gao Gao
Contributing MemberPosts: 545
Like?
Let's make the conditions clearer -
In reply to Silverback1988, May 19, 2012

1 - How does DR change w.r.t. ISO?

For example, with Canon cameras, the change in DR at the lower end is not linear: ISO 100 is only marginally better than 400. With the latest Sony sensors, it seems more or less linear, though.

2 - What kind of resolution are we talking about?

In other words, what is the native contrast of the target, assuming we are using some kind of line pattern similar to the ISO 12233 test chart?

3 - What is the noise reduction technology used?

Linear space-invariant filtering is perhaps one of the most naive way and impact resolution the most. This, however, could serve as a lower bound - meaning, ISO 6400 perhaps offers AT LEAST half the linear resolution of ISO 1600.

With high-contrast, well-structured targets, sophisticated image noise reduction algorithms are able to differentiate signal and noise clearly, and process accordingly. After processing, say, an ISO 1600 image of the 12233 target may well have the same measured resolution as the ISO 100 image. However, this does not mean all natural images will look the same - we have a lot of seemingly random and subtle details submerged by noise and they are difficult to recover with any noise reduction method. The perception here is not a loss of resolution but a somewhat unnatural imbalance in different frequency and contrast components - sometimes as watercolor effect or being plastic/wax-like.

Silverback1988 wrote:

Hi,
I curious if someone of you has been calculate this.

As we know, if we keep the noise constant by applying NR, then there would be a decrease in resolution as we raise the ISO.
But by how much from the native resolution ?

For example we had 20 MP camera.
Say we are able to get the full resolution of 20 MP is on ISO 100.

As we raise the ISO, say ISO 1600, then we keep the noise similar to the ISO 100 by applying NR. How much resolution is left ? Surely is not 20 MP anymore.
Would be it 5 MP only ?

I think we could compare it with a 5 MP camera and then shot with ISO 100.
If the picture is identical then the resolution is the same.

And I think the decrease is not linear between ISO 100 and ISO 400.

Opinions are welcome.

Sorry, English is my third language
I hope I state my question clearly.

Thanks for the reply.

-
Brian

-- hide signature --

. 。o O o 。 . 。o O o 。 . 。o O o 。 .

Reply   Reply with quote   Complain
5tve
Contributing MemberPosts: 638Gear list
Like?
Re: Let's make the conditions clearer -
In reply to Gao Gao, May 19, 2012

I often don't apply any noise reduction to ISO 1600 or even ISO 3200 pics it depends a lot on the the light available when the picture was taken.

For example a photo in good light when you use ISO 3200 to freeze motion in football players is going to have a lot less noise than an ISO 3200 photo of the inside of a Church.

Amateur Photographer mag have reviewed both D800 & 5DMKIII here are the resolution figures reached at the different ISO's
no noise reduction applied though & probably taken in good light

 5tve's gear list:5tve's gear list
Nikon D7000 Sony a6000
Reply   Reply with quote   Complain
bclaff
Contributing MemberPosts: 698
Like?
Re: How much resolution decrease as we raise the ISO ?
In reply to Silverback1988, May 19, 2012

Brian,

I hope you don't mind if I restate your question a little differently.

Rather than resolution, let's talk about how print size of the same quality would vary. This is also equivalent to talking about how much you could crop to get the same print size.

As an example I'll use data on the 5d Mark III from my site:
http://home.comcast.net/~NikonD70/Charts/PDR.htm#EOS%205D%20Mark%20III

First some results, and then the math.
At ISO 100 the PDR is 8.93, let's assume a 18.6x27.9" print.
At ISO 1600 the PDR is 7.50, that's an 11.2x16.8" print of the same quality.

In the above calculations I assumed that a PDR of 6.5 is acceptable quality for an 8" wide print.

So at ISO 1600 the PDR is 7.50 - 6.50 or 1.00 higher. 1.00 stops is 2.00x (2 raised to the 1.00 power) more area or about 1.4x (square root of 2.00) more per side. So 8 * 1.4 = 11.2 and 12 * 1.4 = 16.8
The computation at ISO 100 is similar.

We can also determine how heavily a certain ISO could be cropped and still maintain a minimum print size. If we were targeting an 8x12" print then we could crop the ISO 1600 image to 71% or the ISO 100 image to 43% and produce the same quality 8x12" print.

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

Reply   Reply with quote   Complain
Silverback1988
Contributing MemberPosts: 689
Like?
Wow, thanks b claff.
In reply to bclaff, May 21, 2012

bclaff wrote:

Brian,

I hope you don't mind if I restate your question a little differently.

of course I don't mind at all

Rather than resolution, let's talk about how print size of the same quality would vary. This is also equivalent to talking about how much you could crop to get the same print size.

Okay. I consider it similar (equivalent) to my original question.
That's good.

As an example I'll use data on the 5d Mark III from my site:
http://home.comcast.net/~NikonD70/Charts/PDR.htm#EOS%205D%20Mark%20III

First some results, and then the math.
At ISO 100 the PDR is 8.93, let's assume a 18.6x27.9" print.
At ISO 1600 the PDR is 7.50, that's an 11.2x16.8" print of the same quality.

In the above calculations I assumed that a PDR of 6.5 is acceptable quality for an 8" wide print.

So at ISO 1600 the PDR is 7.50 - 6.50 or 1.00 higher. 1.00 stops is 2.00x (2 raised to the 1.00 power) more area or about 1.4x (square root of 2.00) more per side. So 8 * 1.4 = 11.2 and 12 * 1.4 = 16.8
The computation at ISO 100 is similar.

So, your data shows us that with a 5DIII at ISO 1600, the resolution is decrease to 50% from the native (original) resolution. That would be only about 11 MP.
Is that correct ?

We can also determine how heavily a certain ISO could be cropped and still maintain a minimum print size. If we were targeting an 8x12" print then we could crop the ISO 1600 image to 71% or the ISO 100 image to 43% and produce the same quality 8x12" print.

Wow,
that's a very good information, bclaff.
and a very good analysis too.
Thanks for your effort & sharing.

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

Any body has another finding or data to share ?

Reply   Reply with quote   Complain
bclaff
Contributing MemberPosts: 698
Like?
Re: Wow, thanks b claff.
In reply to Silverback1988, May 21, 2012

Brian,

So, your data shows us that with a 5DIII at ISO 1600, the resolution is decrease to 50% from the native (original) resolution. That would be only about 11 MP.

Is that correct ?
No, lower. The PDR drops from 8.93 to 7.5 that's -1.43 stops or 0.37x; 37%
So it's about 22.1Mp * 37% = 8.2Mp


--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

Reply   Reply with quote   Complain
Danack
Regular MemberPosts: 331
Like?
Interesting charts!
In reply to bclaff, May 21, 2012

One quick question which is not related to the current discussion - on the chart you linked and also this chart:

http://home.comcast.net/~NikonD70/Charts/RN_e.htm

It's pretty obvious that ISO settings not a full stop apart are doing something funny or as you said it "Flat spots on either side of a full ISO indicate digital gain that is present for some Canon models."

From your graphs it looks like there is little to gain from using intermediate ISOs (e.g. ISO 160) and in fact it could be really bad to use those settings, particularly for the 5d original, and the 1DmkIII. Is that a correct interpretation of your data?

cheers
Dan

-- hide signature --
Reply   Reply with quote   Complain
Silverback1988
Contributing MemberPosts: 689
Like?
Re: Wow, thanks b claff.
In reply to bclaff, May 21, 2012

bclaff wrote:
Brian,

So, your data shows us that with a 5DIII at ISO 1600, the resolution is decrease to 50% from the native (original) resolution. That would be only about 11 MP.

Is that correct ?
No, lower. The PDR drops from 8.93 to 7.5 that's -1.43 stops or 0.37x; 37%
So it's about 22.1Mp * 37% = 8.2Mp


--

Thanks for the correction. It's 1.43 stops not 1 stop.

On my original post, I assume the ISO 1600 (from 5DII) is only 5 MP because I feel its IQ just worth around 25% from the native resolution.
That's just purely a guess-work from what I see in my photos.

So now I take your data It is actually 8 MP.
Uupps your data is about 5III, mine is 5DII

Uggh... well let's consider it equal, it's ISO 1600 anyway, not so much between those two

In your chart,
with D800 (ISO 200 until 25600), we are only about 1 stop from ideal FX ?
And how much MP does the ideal FX has ?
2 stops better than D800 will make 2*2*36 = 144 MP.
So I guess the ideal FX is around 100 MP.
200 MP maybe more than ideal. It will gain very little from the ideal one.
That's just a guess-work too

Again,
thanks for the great chart bclaff !
-
Brian

Reply   Reply with quote   Complain
Keyboard shortcuts:
FForum MMy threads