Sigma Noise Comparison: Not Fair part 2

Started Nov 27, 2002 | Discussions
tom jacobson Contributing Member • Posts: 516
Sigma Noise Comparison: Not Fair part 2

I have had a chance to study the noise comparisons more carefully and have found another factor that is throwing the comparisons off. First, as I mentioned in the first thread, Sigma hides or "masks" the noise levels of the higher ISO's by lowering overall contrast and sharpness as you go up in in ISO. This I do not consider "noise reduction", as you will have "noise enhancement" when you correct the image to proper contrast and sharpness. Not an effective way to deal with noise, in my opinion. The second thing that really throws off the comparisons http://www.dpreview.com/reviews/sigmasd9/page15.asp is that we now have the resolution factor throwing things off. If you take a SD-9 patch and interpolate it up by a factor of two, you now have a clumpy noise patch that looks just like the D60, only worse. Phil mentions that the noise of the SD-9 is finer, and guesses that it is because of the non-Bayer pixel setup. In fact, it is just that the noise is not at the same resolution as it is in the D60 patch. If one were to make a D60 patch half as big, or at equal resolution, it would also look very fine, like the SD-9 patch.--
TJ

OP tom jacobson Contributing Member • Posts: 516
Re: Sigma Noise Comparison: Not Fair part 2

I am not suggesting that Phil has rigged things up to make the SD-9 look better than it should. All this is new stuff and mistakes are easy to make. However, it is clear that we are not comparing apples to apples here. One tends to treat the SD-9 in the league of the 6MP cameras. It is in that league, but only if one interpolates down the 6MP images to match the SD-9 ones or visa versa. The comparisons of noise levels http://www.dpreview.com/reviews/sigmasd9/page15.asp are clearly misleading as they were presented. At equal image size, the SD-9 noise will look very much like the D60's. Add contrast and sharpness, and you have a problem!

TJ

tom jacobson wrote:

I have had a chance to study the noise comparisons more carefully
and have found another factor that is throwing the comparisons off.
First, as I mentioned in the first thread, Sigma hides or "masks"
the noise levels of the higher ISO's by lowering overall contrast
and sharpness as you go up in in ISO. This I do not consider "noise
reduction", as you will have "noise enhancement" when you correct
the image to proper contrast and sharpness. Not an effective way
to deal with noise, in my opinion. The second thing that really
throws off the comparisons
http://www.dpreview.com/reviews/sigmasd9/page15.asp is that we now
have the resolution factor throwing things off. If you take a SD-9
patch and interpolate it up by a factor of two, you now have a
clumpy noise patch that looks just like the D60, only worse. Phil
mentions that the noise of the SD-9 is finer, and guesses that it
is because of the non-Bayer pixel setup. In fact, it is just that
the noise is not at the same resolution as it is in the D60 patch.
If one were to make a D60 patch half as big, or at equal
resolution, it would also look very fine, like the SD-9 patch.--
TJ

-- hide signature --

TJ

Herb Ko Regular Member • Posts: 140
Re: Sigma Noise Comparison: Not Fair part 2

There's even bigger problems. Check http://www.dpreview.com/reviews/sigmasd9/page15.asp again, it's been updated.

I've also posted an instance of this problem that I've spotted on one of Phil's test photos at

http://www.dpreview.com/forums/read.asp?forum=1027&message=3851065

OP tom jacobson Contributing Member • Posts: 516
Re: Sigma Noise Comparison: Not Fair part 2

Thanks for pointing that out, Herb.

TJ

Herb Ko wrote:

There's even bigger problems. Check
http://www.dpreview.com/reviews/sigmasd9/page15.asp again, it's
been updated.

I've also posted an instance of this problem that I've spotted on
one of Phil's test photos at

http://www.dpreview.com/forums/read.asp?forum=1027&message=3851065

-- hide signature --

TJ

William Wood Contributing Member • Posts: 778
Re: Sigma Noise Comparison: Not Fair part 2

The "red noise" seems sort of irrelevant since it shows up in the green channel, which is almost dark, and the actual red patch (on page 15) has very similar noise to the D60, despite the high green-channel noise.

Herb Ko wrote:

There's even bigger problems. Check
http://www.dpreview.com/reviews/sigmasd9/page15.asp again, it's
been updated.

I've also posted an instance of this problem that I've spotted on
one of Phil's test photos at

http://www.dpreview.com/forums/read.asp?forum=1027&message=3851065

 William Wood's gear list:William Wood's gear list
Google Pixel 3 Google Pixel 4 XL
Herb Ko Regular Member • Posts: 140
Re: Sigma Noise Comparison: Not Fair part 2

It' s not that different on a bright red patch; it much more noticeable on a dimmer red area, such as the red paddles on that photo on pg 19 of the review. Had you made a large priint of that, I think you would notice the purple spots on those paddles. It's also much more noticeable on the ISO 400 red patch which is really just an amplified version of a lower exposure ASA 100 shot.

BTW. For those who are not on a high speed link and don't want to download the big jpg's, you can see some of what I'm talking about on the red "TLC" sign in on one of the crops. The noise is more noticeable at the red paddles.

William Wood wrote:
The "red noise" seems sort of irrelevant since it shows up in the
green channel, which is almost dark, and the actual red patch (on
page 15) has very similar noise to the D60, despite the high
green-channel noise.

Herb Ko wrote:

There's even bigger problems. Check
http://www.dpreview.com/reviews/sigmasd9/page15.asp again, it's
been updated.

I've also posted an instance of this problem that I've spotted on
one of Phil's test photos at

http://www.dpreview.com/forums/read.asp?forum=1027&message=3851065

-- hide signature --

Herb

Frank C. Veteran Member • Posts: 6,667
Re: Sigma Noise Comparison: Not Fair part 2

tom jacobson wrote:

noise will look very much like the D60's. Add contrast and
sharpness, and you have a problem!

why add sharpness and contrast?..... I mean, the X3 is NOT
a bayer type sensor....

Kok Chen Contributing Member • Posts: 740
Re: Sigma Noise Comparison: Not Fair part 2

By upsampling an image, the signal-to-noise ratio should not appreciably change.

It is true however, that when you do this, the noise turns from the pleasing high frequency noise you see in the example Phil showed, to a more low frequency noise that is more visually objectionable. Perhaps this is what you are alluding to.

The fact that the eye prefers higher frequency noise is pretty well known. The lower frequency noise is sometimes referred to as "pink noise" in the literature (pink has to do with spatial frequency, not the spectral color). And the higher frequency noise is known as "blue noise." That fact that "blue noise" is more pleasing to the human visual system has been used advantageously in halftoning techniques in printing (for example, the Blue Noise Mask). Most of the ink jet printers used in home photo-printing tends to have blue noise characteristics, whether they are directly using the Blue Noise Mask, or whether they are using error diffusion techniques.

Noise from Bayer cameras tend to be "pink" since the samples you see have been interpolated from a sparse array. The interpolation process by necessity is a low pass process, turning what was white noise into pink noise.

  • kc

OP tom jacobson Contributing Member • Posts: 516
Re: Sigma Noise Comparison: Not Fair part 2

I don't think I am getting my point across! Sorry for that. What I am talking about is the size of the noise chunks. The SD-9 noise appears to be fine little chunks, and the D60 appears to have larger chunks of noise. They are not being shown at the same ratio of noise to image output. Think of it a grain. The SD-9 appears to have fine grain, the D60 bigger grain. To make the comparison you should match the output sizes. Does anyone get this???? Maybe I am totally wrong. TJ

Kok Chen wrote:

By upsampling an image, the signal-to-noise ratio should not
appreciably change.

It is true however, that when you do this, the noise turns from the
pleasing high frequency noise you see in the example Phil showed,
to a more low frequency noise that is more visually objectionable.
Perhaps this is what you are alluding to.

The fact that the eye prefers higher frequency noise is pretty well
known. The lower frequency noise is sometimes referred to as "pink
noise" in the literature (pink has to do with spatial frequency,
not the spectral color). And the higher frequency noise is known
as "blue noise." That fact that "blue noise" is more pleasing to
the human visual system has been used advantageously in halftoning
techniques in printing (for example, the Blue Noise Mask). Most of
the ink jet printers used in home photo-printing tends to have blue
noise characteristics, whether they are directly using the Blue
Noise Mask, or whether they are using error diffusion techniques.

Noise from Bayer cameras tend to be "pink" since the samples you
see have been interpolated from a sparse array. The interpolation
process by necessity is a low pass process, turning what was white
noise into pink noise.

  • kc

-- hide signature --

TJ

Kok Chen Contributing Member • Posts: 740
Re: Sigma Noise Comparison: Not Fair part 2

tom jacobson wrote:

I don't think I am getting my point across! Sorry for that. What
I am talking about is the size of the noise chunks. The SD-9 noise
appears to be fine little chunks, and the D60 appears to have
larger chunks of noise.

That is precisely what I was addressing, Tom.

What you refer to as "larger chunks" are simply noise that have a lower spatial frequency structure.

If you add all the "area" of the noise and compare to the area of the rest of the image, that is a kind of crude "signal-to-noise" ratio. This ratio does not change when you scale an image up.

However, the noise will be low-passed (or, as you say, in big chunks) and that, is noise that is more "pink." Again, it is not the color of the noise that is pink, the "pink" refers to lower spatial frequencies. And the human visual system does not like "pink noise."

Call it "small random grain" vs "clumpy noise," if you like.

  • kc

OP tom jacobson Contributing Member • Posts: 516
Re: Sigma Noise Comparison: Not Fair part 2

Thanks.

Kok Chen wrote:

tom jacobson wrote:

I don't think I am getting my point across! Sorry for that. What
I am talking about is the size of the noise chunks. The SD-9 noise
appears to be fine little chunks, and the D60 appears to have
larger chunks of noise.

That is precisely what I was addressing, Tom.

What you refer to as "larger chunks" are simply noise that have a
lower spatial frequency structure.

If you add all the "area" of the noise and compare to the area of
the rest of the image, that is a kind of crude "signal-to-noise"
ratio. This ratio does not change when you scale an image up.

However, the noise will be low-passed (or, as you say, in big
chunks) and that, is noise that is more "pink." Again, it is not
the color of the noise that is pink, the "pink" refers to lower
spatial frequencies. And the human visual system does not like
"pink noise."

Call it "small random grain" vs "clumpy noise," if you like.

  • kc

-- hide signature --

TJ

William Wood Contributing Member • Posts: 778
Re: Sigma Noise Comparison: Not Fair part 2

I don't know. I'd have to see the paddles in person to judge what's going on here. The D60 paddels look articificially red, like they were just painted. The SD9 paddles look rusted, older. Which one is accurate? May not be a noise issue at all. Could be its the D60 Bayer interpolation overstating the red.

Herb Ko wrote:
It' s not that different on a bright red patch; it much more
noticeable on a dimmer red area, such as the red paddles on that
photo on pg 19 of the review. Had you made a large priint of that,
I think you would notice the purple spots on those paddles. It's
also much more noticeable on the ISO 400 red patch which is really
just an amplified version of a lower exposure ASA 100 shot.

BTW. For those who are not on a high speed link and don't want to
download the big jpg's, you can see some of what I'm talking about
on the red "TLC" sign in on one of the crops. The noise is more
noticeable at the red paddles.

William Wood wrote:
The "red noise" seems sort of irrelevant since it shows up in the
green channel, which is almost dark, and the actual red patch (on
page 15) has very similar noise to the D60, despite the high
green-channel noise.

Herb Ko wrote:

There's even bigger problems. Check
http://www.dpreview.com/reviews/sigmasd9/page15.asp again, it's
been updated.

I've also posted an instance of this problem that I've spotted on
one of Phil's test photos at

http://www.dpreview.com/forums/read.asp?forum=1027&message=3851065

 William Wood's gear list:William Wood's gear list
Google Pixel 3 Google Pixel 4 XL
Keyboard shortcuts:
FForum MMy threads