Nikon finally gets the noise down.

Started Nov 17, 2007 | Discussions
PIXmantra Senior Member • Posts: 1,637
That's because...

Amin Sabet wrote:

The way noise and fine details are rendered by LR
at high ISO is just not pleasing to me.

...They are pretty manageable and, ultimately, allow to make very good prints, indeed. LR handles low-frequency noise, but leaves mid and high almost intact, between L=0 to 25, assuming you fine-tune sharpness accordingly.

The results are still a lot more organic-looking (and not detail-robbing) as they are in C1.

I will post further examples on this, with much more detailed images. In any case, FYI, I am anxiously waiting for C1 4.0, in order to see how much have we really advanced here.

I use C1 to
process high ISO 5D files all the time, and that doesn't look right
to me.

...I discovered this A LONG time ago, when playing with CHROMA noise-reduction on C1, and it actually KILLS Luma detail on the shadows. See yourself.

Stay tuned!

-- hide signature --

TIP: If you do not like this post, simply press the 'COMPLAINT' button. Mommy/Daddy are just one click away.

Amin Sabet
Amin Sabet Veteran Member • Posts: 6,763
Re: That's because...

PIXmantra wrote:

Amin Sabet wrote:

The way noise and fine details are rendered by LR
at high ISO is just not pleasing to me.

...They are pretty manageable and, ultimately, allow to make very
good prints, indeed. LR handles low-frequency noise, but leaves mid
and high almost intact, between L=0 to 25, assuming you fine-tune
sharpness accordingly.

The results are still a lot more organic-looking (and not
detail-robbing) as they are in C1.

Disagree that LR results are more organic looking. They look unnatural to me in high ISO images or with small sensor cameras at any ISO. I see it in your examples as well. There are prominent little diagonals in the noise, and there is a "shimmery" appearance overall - I don't know how to describe it well, but "organic" it is not.

I use C1 to
process high ISO 5D files all the time, and that doesn't look right
to me.

...I discovered this A LONG time ago, when playing with CHROMA
noise-reduction on C1, and it actually KILLS Luma detail on the
shadows. See yourself.

I've seen for myself thousands of times what C1 does to files. Any chance you could send me the RAWs to play with? In particular I'd like to try my hand at the one where you circled those regions.

 Amin Sabet's gear list:Amin Sabet's gear list
Sony a9 Panasonic Lumix DMC-GX85 Nikon D500 Nikon D810 Panasonic Lumix G 14mm F2.5 ASPH +16 more
PIXmantra Senior Member • Posts: 1,637
Interesting...

Amin Sabet wrote:

Disagree that LR results are more organic looking. They look
unnatural to me in high ISO images or with small sensor cameras at
any ISO.

...I actually see the opposite: on all my monitors (VP191b, VP930b and EIZO CG241W), I actually see the C1 conversion washed-out and filled with tiny dots (or "artifacts") which are left-overs from its noise-management routines.

I see it in your examples as well. There are prominent
little diagonals in the noise, and there is a "shimmery" appearance
overall

...I can't see the diagonals on any of my monitors (but I can clearly see C1 artifacts, all over). All I see is a noise-spectra that is has a modest low frequency noise impulse, a stronger mid-range one, and an also modest impulse on the high frequency-side. This is a characteristic look, for sure, but clearly more organic/textured than the "wash-machined" version of C1 (my humble opinion). LR will make a better print, especially at higher DPIs.

I've seen for myself thousands of times what C1 does to files.

...Then you may have missed something critical, here, which I noticed when running this comparison, eons ago:

http://www.pbase.com/feharmat/image/46871309/original

On the bottom-center is the C1 sample. I simply had to inhibit CHROMA noise-reduction completely because it would otherwise eat details that were shown on its ISO100 counter-part, or the same 800 sample.

I will try to send the RAW file to you (please, specifiy email address), but, in any case, you will find this immediately with C1, as soon as you shoot something with a good amount of low-mid frequency detail on the SHADOWS.

PIX

-- hide signature --

TIP: If you do not like this post, simply press the 'COMPLAINT' button. Mommy/Daddy are just one click away.

Amin Sabet
Amin Sabet Veteran Member • Posts: 6,763
Re: Interesting...

PIX,

I see the tiny dots you're referring to in C1, but the prominent tiny diagonals and unnatural rendering of microdetail in LR are more bothersome to me. I see what you mean by the "wash-machined" look of C1, and I prefer it to the "overcooked" look of LR. What I find unpleasant, you like. What you see as organic, I see as unnatural. I prefer the prints I get from C1 when dealing with high ISO files, and while I don't print at different DPIs, I do know that the larger I print, the more I tend to prefer C1. I think it all comes down to taste. I believe that many people - most people in fact - would prefer the prints from Lightroom. I'm just not one of those people. The nice thing is, there are tools to suit each of us.

I will try to send the RAW file to you (please, specifiy email
address),

My email address is linked through my profile. Thanks!

Regards,
Amin

 Amin Sabet's gear list:Amin Sabet's gear list
Sony a9 Panasonic Lumix DMC-GX85 Nikon D500 Nikon D810 Panasonic Lumix G 14mm F2.5 ASPH +16 more
photo nuts Senior Member • Posts: 1,364
John, have you seen this?

http://theory.uchicago.edu/~ejm/pix/20d/posts/tests/D300_40D_tests/

Here is a quick summary of the results:

1) The read noise gives a slight edge to the D300 (using extrapolation means... see point 4) at ISO 200, while at higher ISO the 40D has an increasing advantage, up to about 1/4 stop at ISO 1600. Neither camera exhibits significant pattern noise (banding) in deep shadows.
2) Dynamic range values for the D300 are comparable to those of the 40D.

3) D300 sensor has the same quantum efficiency as the 400D, and only slightly behind that of 40D.

4) Nikon clips the black end of the spectrum in the D300 well before zero signal, thereby distorting the noise spectrum in deepest shadows. This precludes any direct measurement of the read noise. The nonlinear distortion of deep shadows in D300 raw data by clipping means that it will be a poor choice for astro-photography, or any application where pulling weak signals out of the noise in at low illumination levels is of importance.

Of course, the above link also shows the superior quantum efficiency of the D3 sensor and the poor performance of the D200/D80/D40x sensor.

OP John Sheehy Forum Pro • Posts: 21,865
Re: I will try to post some samples...

Amin Sabet wrote:

The RAW files themselves have been kindly provided by Pavel Kudrys here -> http://ricohforum.com/phpbb/viewtopic.php?f=39&t=472

Here's a more literal RAW conversion. No NR; only WB and 2.2 gamma, and a little bit of a contrast boost to get the colors at a reasonable saturation:

Comparing this to the JPEGs given, you can see that there is quite a bit of noise here (as there is in any tiny-sensor camera in high-ISO images). All of that scalloped, wormy look in the JPEGs is just a way of tradiing noise into false detail. The blacks are not black in my more literal RAW representation, and that's because of all the noise (about 70 12-bit ADUs of read noise). Blacks are pushed down in typical conversions.

The RAW noise is typical for small sensors.

Once again, the main differences we see in small-sensor camera images is due to software more than it is due to sensor technology. The 6MP Fuji actually has the same or more read noise (depending on ISO) and more shot noise per pixel than the 10MP FZ50, at the RAW level. Through software, Fuji has gained a reputation of a low light wonder, and Panasonic a reputation for not being very usable above ISO 200.

NR is becoming more of a problem than noise, IMO. Other than bringing down the areas that should be black to look almost black, and getting the global color fine-tuned, I'd rather see images looking more like the RAW available from the cameras, at least as an option.

-- hide signature --

John

Amin Sabet
Amin Sabet Veteran Member • Posts: 6,763
John

Thanks for taking a look at those files. I agree with your observations about the RAW image and in-camera JPEGs. Also agree that NR is becoming more of a problem than noise. Do you think that the GRD 2 files have had more NR applied directly to the RAWs as compared with the GRD files?

 Amin Sabet's gear list:Amin Sabet's gear list
Sony a9 Panasonic Lumix DMC-GX85 Nikon D500 Nikon D810 Panasonic Lumix G 14mm F2.5 ASPH +16 more
OP John Sheehy Forum Pro • Posts: 21,865
Re: John

Amin Sabet wrote:

Thanks for taking a look at those files. I agree with your
observations about the RAW image and in-camera JPEGs. Also agree
that NR is becoming more of a problem than noise. Do you think that
the GRD 2 files have had more NR applied directly to the RAWs as
compared with the GRD files?

I don't see any sign of noise-filtering in the GRD 2 ISO 1600 RAW file. It looks pretty natural, with especially bright single pixels that would have been subdued with any kind of filtering.

-- hide signature --

John

Bill Janes Senior Member • Posts: 1,858
Re: How to subtract in quadrature?

A very interesting analysis, but one thing I don't understand is subtracting in quadrature. Can someone explain this?
--
Bill Janes

Amin Sabet
Amin Sabet Veteran Member • Posts: 6,763
Thanks! -nt
 Amin Sabet's gear list:Amin Sabet's gear list
Sony a9 Panasonic Lumix DMC-GX85 Nikon D500 Nikon D810 Panasonic Lumix G 14mm F2.5 ASPH +16 more
PIXmantra Senior Member • Posts: 1,637
Checked those ISO400 files from both (OMG!)

Amin Sabet wrote:

In the meanwhile, I'd like to hear your
thoughts about some ISO 800 crops from a new compact camera, the
Ricoh GRD II. Do you think these crops ->

http://forums.dpreview.com/forums/read.asp?forum=1013&message=25783894 suggest that Ricoh is doing more NR on the RAW files of the GRD II as compared with the GRD? The RAW files themselves have been kindly provided by Pavel Kudrys here -> http://ricohforum.com/phpbb/viewtopic.php?f=39&t=472

...The GRDII seems to be using a combination of a better SENSOR + better signal processing. It is not just one area alone.

Detail is pretty well handled (and in fact easier to see) on the GRDII shot, when compared to GRD. Also, GRDII noise-spectra seems a lot more manageable than GRD, which has a pretty stronger low-frequency impulse (much more "chunkier" than GRDII).

Noise on both is equivalent to ISO3200 and ISO6400 on our 1D Mark III, though.

My initial comments, and maybe samples later.

PIX

-- hide signature --

TIP: If you do not like this post, simply press the 'COMPLAINT' button. Mommy/Daddy are just one click away.

Amin Sabet
Amin Sabet Veteran Member • Posts: 6,763
Thanks PIX

Appreciate your thoughts on those files!

 Amin Sabet's gear list:Amin Sabet's gear list
Sony a9 Panasonic Lumix DMC-GX85 Nikon D500 Nikon D810 Panasonic Lumix G 14mm F2.5 ASPH +16 more
alanr0 Senior Member • Posts: 2,203
Re: How to subtract in quadrature?

Bill Janes wrote:

A very interesting analysis, but one thing I don't understand is
subtracting in quadrature. Can someone explain this?

The assumption is that photon noise (P) and readout noise (R) are uncorrelated. In this case the total noise variance is the sum of the variances of photon and readout noises. The RMS noise (N) is therefore given by the square root of the sum of variances. The two components add in quadrature.
N = sqrt(P^2 + R^2)

We need the photon noise variance to estimate the quantum efficiency. We estimate this by subtracting the read-out variance from the total noise variance. This is what is meant by subtracting in quadrature.
P^2 = N^2 - R^2

HTH
--
Alan Robinson

PIXmantra Senior Member • Posts: 1,637
UPDATE: Ricoh's GRD2's NR-engine is similar to D3!!!

...I will be posting my subjective analysis, with samples to illustratre, pretty soon.

Stay tuned.

PIX

-- hide signature --

TIP: If you do not like this post, simply press the 'COMPLAINT' button. Mommy/Daddy are just one click away.

Bill Janes Senior Member • Posts: 1,858
Re: Thanks Alan

I should have been able to figure that out for myself, but your explanation was excellent.

Regards,

Bill Janes

Eric Sorensen
Eric Sorensen Veteran Member • Posts: 4,641
Collect photons twice as fast? I don't get it.

Please explain this - I'm confused.

-- hide signature --

Bossier City, Louisiana
http://www.pbase.com/ericsorensen

 Eric Sorensen's gear list:Eric Sorensen's gear list
Canon EOS 5D Mark II Canon EOS 5D Canon EF 85mm f/1.2L II USM Tamron SP 24-70mm F2.8 Di VC USD Sigma 12-24mm F4.5-5.6 EX DG Aspherical HSM +3 more
Don_D
Don_D Forum Pro • Posts: 20,658
Re: Collect photons twice as fast? I don't get it.

Eric Sorensen wrote:

Please explain this - I'm confused.

It has been determined by John and others that the D3's sensor has a 2x higher quantum efficiency than the 5D's sensor. That is, for each photon that strikes the D3's sensor, twice number of electron hole pairs, and therefore twice the charge is generated compared to the 5D's sensor.
I think I have this right..if not someone can please correct me.
Don
http://www.pbase.com/dond

 Don_D's gear list:Don_D's gear list
Sony RX100 Canon EOS 40D Canon EF 70-200mm F4L USM Canon EF-S 15-85mm f/3.5-5.6 IS USM Canon EF-S 17-55mm f/2.8 IS USM +3 more
PIXmantra Senior Member • Posts: 1,637
Here we go (ALL samples)...

First, ISO400 from GRD and GRD2, left and right respectively, and best I could from Lightroom without killind detail and using any external tool, whatsoever(notice a complete 1-to-1.5 stop advantage of GRD2 over its former sibbling, and that is wit less NR applied to GRD2 in Lightroom):

http://www.pbase.com/feharmat/image/89617775/original
http://www.pbase.com/feharmat/image/89617776/original

Please, note that these images are really NOISY on their natural form, and without any NR in Lightroom.

Enter the .JPG vs RAW comparison... WOW!!!

I must say that this little cam has a true NR workhorse on-board that literally TRANSFORMS these images into something different. However, there are certainly side effects (pretty similiar to those I see on D3 .JPGs). Here is the .JPG version from RD2 (left) and my best-non-destructive version from Lightroom (RAW) on the right:

http://www.pbase.com/feharmat/image/89617778/original
http://www.pbase.com/feharmat/image/89617779/original

Now, on the last image, and the isolated area on the left (.JPG) that has the "D3" label and compare to this sequence from D3 (also with "pasta" as being the subject) and notice the strikingly similar look on the ISO6400 clip:

http://www.pbase.com/feharmat/image/89084939/original

All I can say is that it is now evident HOW MUCH you can really improve an image's subjective quality by having such kind of image processing, which in the case of the Ricoh GRD2 is happening off-sensor, without any doubts.

PIX

-- hide signature --

TIP: If you do not like this post, simply press the 'COMPLAINT' button. Mommy/Daddy are just one click away.

PerL Forum Pro • Posts: 14,052
Re: I finally understand Phil's rating

photo nuts wrote:

for image quality of 40D (9 out of 10).

From http://www.astrosurf.com/buil/nikon_test/test.htm

"According the current measures, the Canon 40D CMOS is slightly less
noisy than the CMOS Nikon D3 CMOS, but the gap is not very
significant......

The readout noise of the Nikon and Canon CMOS detectors is very
similar. The high ISO sensitivity displayed by Nikon is for the
marketing: Nikon D3 CMOS (and also the CMOS D300 CMOS - Sony origin)
is a very high quality sensor, but in the end, Nikon just rises now
to the level of Canon technology."

So, if Canon has implemented a more aggressive noise-reduction
(fortunately NOT) like Nikon, the 40D may very well have the same
performance as the D3? Ho, ho, ho....

Which the 40D obviously not has, no matter how much you work with NR, so something is probably not right.

In this analysis the D3 tops every quality parameter regarding quantum efficiency, light collecting ability, etc. Also interesting to note is that the D300 seems to be a 1/3 stop more sensitive than the 40D, it has been the other way around before.
http://theory.uchicago.edu/~ejm/pix/20d/posts/tests/D300_40D_tests/
--
Small D200 gallery:
http://www.pbase.com/interactive/d200_12
Small D40 gallery:
http://www.pbase.com/interactive/d40_12
Small Nikon P5000 gallery:
http://www.pbase.com/interactive/nikonp5000_12
http://www.pbase.com/interactive

PIXmantra Senior Member • Posts: 1,637
Just a pity, thought, that...

PerL wrote:

photo nuts wrote:
In this analysis the D3 tops every quality parameter regarding
quantum efficiency, light collecting ability, etc.

...That is at ISO400. As wel all know, that will substantially change as you move along the ISO scale. As an example, the 5D tops out at 80K electrons (quite more than the D3 at its baseline ISO200 speed) at ISO100, while the 1D2/N tops out at 80K electrons at ISO75 (quite more than D3 at its max. reading at ISO200) which you access by setting CF for ISO expansion and dialing in a permanent -0.3EV E.C.).

Also interesting
to note is that the D300 seems to be a 1/3 stop more sensitive than
the 40D, it has been the other way around before.

...A pity that both D3 and D300 will not be able to be used on AstroPhotrography, though, because they kill/wash the RAW file on long-time exposures, and they also clip the blacks. Now, if Nikon decides to add parameter on C. Functions, for inhibiting such cheating, then it would be a WHOLE different story.

PIX

-- hide signature --

TIP: If you do not like this post, simply press the 'COMPLAINT' button. Mommy/Daddy are just one click away.

Keyboard shortcuts:
FForum MMy threads