From DXOMark: More pixels offsets noise3

Started May 30, 2009 | Discussions
berleconi Contributing Member • Posts: 720
From DXOMark: More pixels offsets noise3

The original question of before thread was:

just noticed this:
from dxomark site

http://www.dxomark.com/index.php/eng/Insights/More-pixels-offsets-noise !

"Contrary to conventional wisdom, higher resolution actually compensates for noise"
As DPreview does not wants to change the 100% pixel view
Because it is there a very long time.
Maybe we can ask for a extension as was put forward by experts
For example upscale to 100MP and then show a crop

Not a full scale comparometer only a simple crop else it uses
To maybe great a bandwidth

The same size of the crop the same percent of the sensor area

Berl.
http://FreePhotoSoftware.shorturl.com

Steen Bay Veteran Member • Posts: 7,418
Bob..

Just one question. You said in the other thread that a gaussian blur before downsampling wouldn't affect detail/sharpness (and Emil said that it would reduce noise in the downsampled image), but you also said in a reply to zxaar that "you cannot know which pixels are 'noisy' that is the nature of noise". So, how can gaussian blur remove noise, without removing detail at the same time?

And btw, Phil's reply is still there, in the first thread :

http://forums.dpreview.com/forums/read.asp?forum=1018&message=31922021

OP berleconi Contributing Member • Posts: 720
Bob..

Bob can you explain the diffraction
And what it does with pixels smaller than the diffraction limit?
I understand the aliasing but not diffraction
Because my computer program shows that waves have no problems
Entering small holes
Is it the micro lenses that create a limit?
Berl.

http://FreePhotoSoftware.shorturl.com

bobn2
bobn2 Forum Pro • Posts: 61,416
Re: Bob..

Steen Bay wrote:

Just one question. You said in the other thread that a gaussian blur
before downsampling wouldn't affect detail/sharpness (and Emil said
that it would reduce noise in the downsampled image), but you also
said in a reply to zxaar that "you cannot know which pixels are
'noisy' that is the nature of noise". So, how can gaussian blur
remove noise, without removing detail at the same time?

Emil explained this nicely in his post on the subject. Remember, you are reducing detail/sharpness by downsampling. If you do that with high frequency noise still there, that noise aliases into the downsampled image, so before you downsample you blur which averages the detail and noise above the maximum spatial frequency to be present in the output image. The blur is a crude low pass filter, it eliminates noise and detail, but only that which will not appear in the output image, due to the lower sampling rate. Since you've taken away the high frequency noise, it can't be aliased into the lower resolution image. Not doing the blur was the mistake PIX was making, and his downsamples showed that aliased high frequency noise.

-- hide signature --

Bob

bobn2
bobn2 Forum Pro • Posts: 61,416
Re: Bob..

berleconi wrote:

Bob can you explain the diffraction
And what it does with pixels smaller than the diffraction limit?
I understand the aliasing but not diffraction
Because my computer program shows that waves have no problems
Entering small holes

Not quite sure what you mean by that.

Is it the micro lenses that create a limit?

No, it's the taking lens. Every lens is diffraction limited. Diffraction gives a charcteristic MTF contrast/frequency curve, which is a low pass function, hitting 0 contrast at some spatial frequency. The pixel pitch defines a sampling frequency, and once that is at or above twice the diffraction limit (taking that as the zero contrast point) there is no more information to be had from that system - you can decrease the sample as much as you like, and no more information will be captured. In practice, the 'limit' would be set at a higher contrast than zero, and we might exclude some esoteric and excellent lenses, but when the sampling frequency is above the maximum physically possible resolution of any lens, there is no point having an AA filter.

The effect of the microlenses is to change the sample from a discrete sample to an averaged sample over the width of the pixel. This in itself will reduce the very highest frequency response of the system, another factor that oversampling can overcome.
--
Bob

Kuivaamo Senior Member • Posts: 2,248
Re: Bob..

bobn2 wrote:

In practice, the 'limit' would be set at a higher
contrast than zero, and we might exclude some esoteric and excellent
lenses, but when the sampling frequency is above the maximum
physically possible resolution of any lens, there is no point having
an AA filter.

If this approach is eventually taken, let's hope Canon won't release a diffraction-limited 50/1.2; by my quick and dirty estimate we'd need a 950 megapixel APS-C sensor to avoid aliasing without an AA filter.

bobn2
bobn2 Forum Pro • Posts: 61,416
Re: Bob..

Kuivaamo wrote:

bobn2 wrote:

In practice, the 'limit' would be set at a higher
contrast than zero, and we might exclude some esoteric and excellent
lenses, but when the sampling frequency is above the maximum
physically possible resolution of any lens, there is no point having
an AA filter.

If this approach is eventually taken, let's hope Canon won't release
a diffraction-limited 50/1.2; by my quick and dirty estimate we'd
need a 950 megapixel APS-C sensor to avoid aliasing without an AA
filter.

True. Thats why I excluded esoteric and excellent lenses. Generally lenses are aberration limited at wide apertures, and so far as I know, none even approaches the theoretical diffraction limit at f/4 or so. At that point, aberrations are your friend. I would have thought 500 pixels/mm (216MP on FF) would be enough to see pretty much every lens below 10% contrast over its whole aperture range, and that's within the range of current P&S pixels. Could be wrong, though.

-- hide signature --

Bob

Kuivaamo Senior Member • Posts: 2,248
Re: Bob..

bobn2 wrote:

Kuivaamo wrote:

bobn2 wrote:

In practice, the 'limit' would be set at a higher
contrast than zero, and we might exclude some esoteric and excellent
lenses, but when the sampling frequency is above the maximum
physically possible resolution of any lens, there is no point having
an AA filter.

If this approach is eventually taken, let's hope Canon won't release
a diffraction-limited 50/1.2; by my quick and dirty estimate we'd
need a 950 megapixel APS-C sensor to avoid aliasing without an AA
filter.

True. Thats why I excluded esoteric and excellent lenses. Generally
lenses are aberration limited at wide apertures, and so far as I
know, none even approaches the theoretical diffraction limit at f/4
or so. At that point, aberrations are your friend. I would have
thought 500 pixels/mm (216MP on FF) would be enough to see pretty
much every lens below 10% contrast over its whole aperture range, and
that's within the range of current P&S pixels. Could be wrong, though.

You're probably correct, or close to it anyway. Most DSLR lenses, even very good ones, seem to be mostly aberration-limited even at f8. A truly diffraction-limited lens at f8 should comfortably "outresolve" the 50D sensor, but I haven't seen any tests where this is the case.

DSPographer Senior Member • Posts: 2,464
Diffraction cutoff frequency

Kuivaamo wrote:

bobn2 wrote:

In practice, the 'limit' would be set at a higher
contrast than zero, and we might exclude some esoteric and excellent
lenses, but when the sampling frequency is above the maximum
physically possible resolution of any lens, there is no point having
an AA filter.

If this approach is eventually taken, let's hope Canon won't release
a diffraction-limited 50/1.2; by my quick and dirty estimate we'd
need a 950 megapixel APS-C sensor to avoid aliasing without an AA
filter.

-- hide signature --

The commonly cited diffraction limit is much coarser than the diffraction cutoff frequency where contrast goes to zero. The pixel spacing for that limit is wavelength*f-stop/2. So for .55 micron green light in the middle of the spectrum at f1.2 the spacing for zero aliasing because of diffraction is .33 microns which is equivalent to a 3 Gigapixel APS-C sensor. The ten percent contrast point would be at .4 micron pixel pitch. This doesn't include the fact that for color Bayer pattern filters the spacing would need to be even more fine to avoid color aliasing. I don't expect to see real f1.2 lenses that have much contrast near the diffraction cutoff but f2.8 lenses can be designed with fairly low aberrations in the image center.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
OP berleconi Contributing Member • Posts: 720
Re: Bob..

But .microscopy lenses do
What is the F number of a NA 0.95 microscope objective for example.
I am a bit lazy to calculate that now. F 0.3 ??
And lines MM-- how exotic can it be in the future?
Berl.

http://FreePhotoSoftware.shorturl.com

DSPographer Senior Member • Posts: 2,464
Re: Bob..

Kuivaamo wrote:

You're probably correct, or close to it anyway. Most DSLR lenses,
even very good ones, seem to be mostly aberration-limited even at f8.
A truly diffraction-limited lens at f8 should comfortably
"outresolve" the 50D sensor, but I haven't seen any tests where this
is the case.

The term outresolve is not really appropriate. The system MTF is the product of the Lens MTF and the focus-error/motion-blur MTF and the sensor/anti-aliasing filter MTF. The definition of how high a frequency can be resolved also depends on what contrast level you place the threshold at and how much sharpening you apply before the measurement. I own three lenses that I expect would easily benefit from a pixel pitch much finer than the 50D's over the f4-f8 range at the image center providing meticulous technique is used: my 50/1.8, 100/2.8 macro and 200/2.8L.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
Kuivaamo Senior Member • Posts: 2,248
Re: Bob..

DSPographer wrote:

Kuivaamo wrote:

You're probably correct, or close to it anyway. Most DSLR lenses,
even very good ones, seem to be mostly aberration-limited even at f8.
A truly diffraction-limited lens at f8 should comfortably
"outresolve" the 50D sensor, but I haven't seen any tests where this
is the case.

The term outresolve is not really appropriate. The system MTF is the
product of the Lens MTF and the focus-error/motion-blur MTF and the
sensor/anti-aliasing filter MTF. The definition of how high a
frequency can be resolved also depends on what contrast level you
place the threshold at and how much sharpening you apply before the
measurement. I own three lenses that I expect would easily benefit
from a pixel pitch much finer than the 50D's over the f4-f8 range at
the image center providing meticulous technique is used: my 50/1.8,
100/2.8 macro and 200/2.8L.

I'm aware of all this, hence me placing "outresolve" in inverted commas. I would expect much higher lp/mm results in all the lens tests if lenses really were diffraction-limited at f8. There's no reason why an aberration-free optic would resolve less at f8 than f5.6 on a 50D; with a high contrast target the 50D can resolve only about 70% of what I'd expect an ideal lens to deliver at f8.

OP berleconi Contributing Member • Posts: 720
Re: Bob..

Sorry my misunderstanding of a earlier post.

Where was talk about pixels and diffraction limit

So I mixed up pixels with the lens
Berl.
http://FreePhotoSoftware.shorturl.com

foot Veteran Member • Posts: 3,897
how about a concrete example?

i keep hearing ppl trash-talk pixel-peeping, but they offer nothing better. Upsizing, downsizing, blur before downsizing....everyting but an example to make your point.

so why not come up with an example test case to make your point?

imaging resources has lots of images from lots of cameras.
http://www.imaging-resource.com/IMCOMP/COMPS01.HTM

 foot's gear list:foot's gear list
Sigma dp2 Quattro
Steen Bay Veteran Member • Posts: 7,418
Re: Bob..

bobn2 wrote:

Steen Bay wrote:

Just one question. You said in the other thread that a gaussian blur
before downsampling wouldn't affect detail/sharpness (and Emil said
that it would reduce noise in the downsampled image), but you also
said in a reply to zxaar that "you cannot know which pixels are
'noisy' that is the nature of noise". So, how can gaussian blur
remove noise, without removing detail at the same time?

Emil explained this nicely in his post on the subject. Remember, you
are reducing detail/sharpness by downsampling. If you do that with
high frequency noise still there, that noise aliases into the
downsampled image, so before you downsample you blur which averages
the detail and noise above the maximum spatial frequency to be
present in the output image. The blur is a crude low pass filter, it
eliminates noise and detail, but only that which will not appear in
the output image, due to the lower sampling rate. Since you've taken
away the high frequency noise, it can't be aliased into the lower
resolution image. Not doing the blur was the mistake PIX was making,
and his downsamples showed that aliased high frequency noise.

Bob

Thanks. Reading Emil's reply again, I can see that he actually did explain it nicely (and you too, of course), but still I find it a bit counter-intuitive, that a gaussian blur affects the visible noise in the downsampled image, but not the visible detail. Anyway, I'll take your word for it, that it's the best method when downsampling, also to prevent aliasing. (And it would of course be pretty easy to check, downsampling the 50D resolution testchart shot, with and without the gaussian blur)

bobn2
bobn2 Forum Pro • Posts: 61,416
Re: Bob..

Here's one I prepared earlier.
40D and 50D, processed in dcraw at defaults, no other processing.

Rerezzed to 30MP,22MP,15MP,10MP,6.7MP. Upsampling straight Lanczos. Downsampling appropriate blur + Lanczos. No other processing applied. Sorry for just 200x200 crops, makes it faster to view in DPR

Steen Bay wrote:

bobn2 wrote:

Steen Bay wrote:

Just one question. You said in the other thread that a gaussian blur
before downsampling wouldn't affect detail/sharpness (and Emil said
that it would reduce noise in the downsampled image), but you also
said in a reply to zxaar that "you cannot know which pixels are
'noisy' that is the nature of noise". So, how can gaussian blur
remove noise, without removing detail at the same time?

Emil explained this nicely in his post on the subject. Remember, you
are reducing detail/sharpness by downsampling. If you do that with
high frequency noise still there, that noise aliases into the
downsampled image, so before you downsample you blur which averages
the detail and noise above the maximum spatial frequency to be
present in the output image. The blur is a crude low pass filter, it
eliminates noise and detail, but only that which will not appear in
the output image, due to the lower sampling rate. Since you've taken
away the high frequency noise, it can't be aliased into the lower
resolution image. Not doing the blur was the mistake PIX was making,
and his downsamples showed that aliased high frequency noise.

Bob

Thanks. Reading Emil's reply again, I can see that he actually did
explain it nicely (and you too, of course), but still I find it a bit
counter-intuitive, that a gaussian blur affects the visible noise in
the downsampled image, but not the visible detail. Anyway, I'll take
your word for it, that it's the best method when downsampling, also
to prevent aliasing. (And it would of course be pretty easy to check,
downsampling the 50D resolution testchart shot, with and without the
gaussian blur)

-- hide signature --

Bob

bobn2
bobn2 Forum Pro • Posts: 61,416
Re: Bob..

Steen Bay wrote:

Thanks. Reading Emil's reply again, I can see that he actually did
explain it nicely (and you too, of course), but still I find it a bit
counter-intuitive, that a gaussian blur affects the visible noise in
the downsampled image, but not the visible detail. Anyway, I'll take
your word for it, that it's the best method when downsampling, also
to prevent aliasing. (And it would of course be pretty easy to check,
downsampling the 50D resolution testchart shot, with and without the
gaussian blur)

Here you are 50D downsampled to 6.7MP, this time cubic, top no blur, bottom blur.

-- hide signature --

Bob

bobn2
bobn2 Forum Pro • Posts: 61,416
One last one...

Just to illustrate:
Top one is 50D Lanczos resampled to 30MP
Second is 50D, blurred 0.7 pixel radius and Lanczos resampled to 30MP

Third is 50D, blurred 0.7 pixel radius, Lanczos resampled to 10MP then Lanczos resampled to 30MP
Fourth is 40D, Lanczos resampled to 30MP.

-- hide signature --

Bob

Steen Bay Veteran Member • Posts: 7,418
Re: Bob..

bobn2 wrote:

Steen Bay wrote:

Thanks. Reading Emil's reply again, I can see that he actually did
explain it nicely (and you too, of course), but still I find it a bit
counter-intuitive, that a gaussian blur affects the visible noise in
the downsampled image, but not the visible detail. Anyway, I'll take
your word for it, that it's the best method when downsampling, also
to prevent aliasing. (And it would of course be pretty easy to check,
downsampling the 50D resolution testchart shot, with and without the
gaussian blur)

Here you are 50D downsampled to 6.7MP, this time cubic, top no blur,
bottom blur.

Bob

Interesting, thanks. Isn't the top one, with no gaussian blur, the sharpest, with most detail and noise? Does it have any aliasing/artifacts? ISO 3200?

bobn2
bobn2 Forum Pro • Posts: 61,416
Re: Bob..

Steen Bay wrote:

bobn2 wrote:

Steen Bay wrote:

Thanks. Reading Emil's reply again, I can see that he actually did
explain it nicely (and you too, of course), but still I find it a bit
counter-intuitive, that a gaussian blur affects the visible noise in
the downsampled image, but not the visible detail. Anyway, I'll take
your word for it, that it's the best method when downsampling, also
to prevent aliasing. (And it would of course be pretty easy to check,
downsampling the 50D resolution testchart shot, with and without the
gaussian blur)

Here you are 50D downsampled to 6.7MP, this time cubic, top no blur,
bottom blur.

Bob

Interesting, thanks. Isn't the top one, with no gaussian blur, the
sharpest, with most detail and noise? Does it have any
aliasing/artifacts? ISO 3200?

Could be I've overdone the blur a tad - if I got it right it shouldn't affect sharpness. Could well be aliasing in the top one, you'd need to hunt around the fabrics.
These are all ISO3200, same CR2's as PIX used.
--
Bob

Keyboard shortcuts:
FForum MMy threads