High ISO raw comparation k20/40d/d300/d3

Started Jul 9, 2008 | Discussions
interpix Regular Member • Posts: 387
High ISO raw comparation k20/40d/d300/d3

ISO 1600 shots from Imaging resource. Raw files converted without noise reduction and sharpening in Raw therapee. 100% crops resized to be the same.

K20D

40D

K20D

40D

K20D

40D

K20D/D300 iso 1600 and D3 at iso 3200

Oleg_V Contributing Member • Posts: 627
Re: High ISO raw comparation k20/40d/d300/d3

Not very fair comparison since in case of K20 NR is applied to RAW files at ISO 1600 and above.

Roland Mabo Forum Pro • Posts: 12,462
Re: High ISO raw comparation k20/40d/d300/d3

All those cameras has Noise Reduction on the sensor that affects the RAW output.

For JPEG's, it is said (vaguely) in the manual that the K20D applies NR at high ISO (at 1600 and ebove) even when the NR is set to off - but several tests has showned that this is not correct. It is a difference between NR on/off at high ISO settings in JPEG's on the K20D. (this is for the NR algorithms done by the PRIME processor).

But this doesn't matter for this test since this is for RAW, where all cameras featured in the comparision do NR on the sensor, and this affects RAW. (the NR algorithms in the PRIME processor is for JPEG's, and does not affect RAW).

So the comparision is very valid.
--
Take care
R
http://www.flickr.com/photos/raphaelmabo

alanschamber Senior Member • Posts: 1,200
So What???

It was not fair either for the A700... but everyone did it the very same way. Even... don't you think that the D300 has noise reduction applied at RAW level?

Yep... it does. CMOS do that... and that's why Pentax headed to CMOS (they didn't need the speed, because of only 3fps).
Canon does the very same thing... and no one says it's not fair.

I think it's not only fair, but it is also worth the comparison.
Alan.
--
Progress is not possible without deviation from the norm - Frank Zappa

Oleg_V Contributing Member • Posts: 627
Re: So What???

alanschamber wrote:

don't you think that the D300 has noise reduction
applied at RAW level?

No

Canon does the very same thing... and no one says it's not fair.

There is common misunderstanding. I call NR the procedure involving readout from several adjacent pixels so it is killing details. Canon don't use such algorithms in RAW, its on chip circuit is just remove dark noise on every single pixel not affecting details.

Oleg_V Contributing Member • Posts: 627
Re: High ISO raw comparation k20/40d/d300/d3

Roland Mabo wrote:

All those cameras has Noise Reduction on the sensor that affects the
RAW output.

Only K20 has NR affecting details no matter which NR option you set. Sony A700 is same sure ..

OP interpix Regular Member • Posts: 387
some more iso 1600

I dont know if K20 puts more or any NR on raw files but here are 2 of my own shots magnified trying to compare details. To the left k20 (Rawtherapee DA 16-50/2.8) and to the right 40D( Lightroom 17-55/2.8)

It seems like the color temperature of the light affects noise levels, with varmer light better for the Canon creating compareble less noise in blue channel with K20D. This is a daylight shot.

OP interpix Regular Member • Posts: 387
ISO 3200

It does not look very smeared to me.

Roland Mabo Forum Pro • Posts: 12,462
Re: So What???

You are misinformed. Nikon applies on-chip NR just as Canon and Pentax are doing.

Pentax uses on-chip NR circuits. The NR algorithms in the PRIME-processor does not affect the RAW output, only the JPEG's. The exact details of the on-chip NR circuits for the Pentax/Samsung CMOS processor is unknown, Samsung has only presented a vague material, but it seems to be based upon same ideas and concept as the on-chip NR used by Canon.
--
Take care
R
http://www.flickr.com/photos/raphaelmabo

Roland Mabo Forum Pro • Posts: 12,462
Re: High ISO raw comparation k20/40d/d300/d3

Oleg_V wrote:

Only K20 has NR affecting details no matter which NR option you set.

It is true that NR on chip reduces some details, but this is also true for Canon and Nikon which uses similar technologies.

How can you say that K20D reduces details more than competition when this on-chip NR can't be turned off, so you can't really know how much the on-chip NR reduces or not since it can be compared with the same processor witout on-chip NR because the NR reduction is part of the sensor design and can't be removed. It is the same story with Canon and Nikon CMOS.

The NR algorithms applied by the PRIME processor for JPEG's destroys details much more, but this can be turned on/off - and this does not affect the RAW files.

What can be said is that the Adobe solutions for RAW, that is the "standard" many compares too, are not ideally suited for the K20D. Different RAW-converters makes different jobs. RAW-therapee seems to do a better job at converting K20D RAW-files than Adobes solution does.
--
Take care
R
http://www.flickr.com/photos/raphaelmabo

Oleg_V Contributing Member • Posts: 627
Re: So What???

Roland Mabo wrote:

You are misinformed.

I have my own RAW files analyzer that can detect NR in case it's done in camera.
So I trust it more than what you have wrote.

Oleg_V Contributing Member • Posts: 627
Re: High ISO raw comparation k20/40d/d300/d3

Roland Mabo wrote:

How can you say that K20D reduces details more than competition when
this on-chip NR can't be turned off, so you can't really know how
much the on-chip NR reduces

In fact I can - just by measuring noise correlations. Without NR the noise is uncorrelated.

GordonBGood Veteran Member • Posts: 6,311
Re: High ISO raw comparation k20/40d/d300/d3

Oleg_V wrote:

Roland Mabo wrote:

How can you say that K20D reduces details more than competition when
this on-chip NR can't be turned off, so you can't really know how
much the on-chip NR reduces

In fact I can - just by measuring noise correlations. Without NR the
noise is uncorrelated.

Hi Oleg, I affirm that you can tell. After all, I checked your math

Best regards, GordonBGood

Oleg_V Contributing Member • Posts: 627
Re: High ISO raw comparation k20/40d/d300/d3

Thank you, Gordon!

I'm not going to say that K20 is awful camera but I'd personally prefer unprocessed RAW.

jl_smith Veteran Member • Posts: 3,917
Re: High ISO raw comparation k20/40d/d300/d3

Hey everyone, let's read the manual.

Page 78:

"Noise Reduction is set to [Strong] when shooting with a sensitivity of ISO 3200 or higher, regardless of the [18. High-ISO Noise Reduction] setting (p.80) in the [C Custom Setting] menu."

Not at ISO1600.
Hurray for getting the facts straight!

 jl_smith's gear list:jl_smith's gear list
Sony RX10 IV Fujifilm X-E3
GordonBGood Veteran Member • Posts: 6,311
Re: High ISO raw comparation k20/40d/d300/d3

jl_smith wrote:

Hey everyone, let's read the manual.

Page 78:

"Noise Reduction is set to [Strong] when shooting with a sensitivity
of ISO 3200 or higher, regardless of the [18. High-ISO Noise
Reduction] setting (p.80) in the [C Custom Setting] menu."

Not at ISO1600.
Hurray for getting the facts straight!

No, that's the JPEG processing high ISO NR that the manual is talking about and that doesn't affect the raw files no matter where it is set. What we are talking about here is NR applied to the raw files that can't be turned off for ISO's of 1600 and higher.

As Oleg says, it only reduces detail levels a little bit at the same time as it reduces noise and doesn't totally destroy the images, but raw purists would rather it wasn't there.

Regards, GordonBGood

jl_smith Veteran Member • Posts: 3,917
Re: High ISO raw comparation k20/40d/d300/d3

GordonBGood wrote:

jl_smith wrote:

Hey everyone, let's read the manual.

Page 78:

"Noise Reduction is set to [Strong] when shooting with a sensitivity
of ISO 3200 or higher, regardless of the [18. High-ISO Noise
Reduction] setting (p.80) in the [C Custom Setting] menu."

Not at ISO1600.
Hurray for getting the facts straight!

No, that's the JPEG processing high ISO NR that the manual is talking
about and that doesn't affect the raw files no matter where it is
set. What we are talking about here is NR applied to the raw files
that can't be turned off for ISO's of 1600 and higher.

As Oleg says, it only reduces detail levels a little bit at the same
time as it reduces noise and doesn't totally destroy the
images, but raw purists would rather it wasn't there.

Regards, GordonBGood

I would love to read actual proof of what you say. Where is this information coming from? Because it's certainly not coming from the product manual. The blurb of ISO3200+ = High NR says absolutely nothing about whether this is only applied to RAW or JPEG shots.

Seriously - If someone can produce a link I would love to read it.

 jl_smith's gear list:jl_smith's gear list
Sony RX10 IV Fujifilm X-E3
Roland Mabo Forum Pro • Posts: 12,462
Re: High ISO raw comparation k20/40d/d300/d3

When the Samsung/Pentax sensor was released, we had many links here to images and information in Korean-language about the basic concept of the on-chip NR technology that is used on the sensor. (this seems to have been part of the official Korean press kit, I have not seen it available in English). The information was vague and lacked many in-depth technical details but it did gave the impression that it was similar to Canon's on-chip NR technology.

I have not seen any official information that confirms Oleg's statements.
--
Take care
R
http://www.flickr.com/photos/raphaelmabo

Roland Mabo Forum Pro • Posts: 12,462
Re: So What???

But can you compare it with the sensor output without the on-chip NR system?

If not, how can you tell that the details are lost more with the Pentax/Samsung sensor than the competition? I don't believe one can tell the output from a sensor without on-chip NR from a RAW file with on-chip NR. You must have a file without on-chip NR to compare with. And this is not possible in this case because the NR is on-chip and can't be removed from the sensor.

No one is denying that the Samsung/Pentax CMOS has on-chip NR system, just as other CMOS sensors has, and this does affect the RAW signal. It must be since it is on-chip before the image processing unit.

I therefore does not see the comparision as being unfair because the CMOS competition also uses on-chip NR and you can't turn it off, they all have it. So there is no other choices to RAW files here, you get what you get.

I have not seen evidence that the PRIME processor uses NR-algorithms that affects the RAW output.
--
Take care
R
http://www.flickr.com/photos/raphaelmabo

GordonBGood Veteran Member • Posts: 6,311
Re: High ISO raw comparation k20/40d/d300/d3

jl_smith wrote:

GordonBGood wrote:

jl_smith wrote:

Hey everyone, let's read the manual.

Page 78:

"Noise Reduction is set to [Strong] when shooting with a sensitivity
of ISO 3200 or higher, regardless of the [18. High-ISO Noise
Reduction] setting (p.80) in the [C Custom Setting] menu."

Not at ISO1600.
Hurray for getting the facts straight!

No, that's the JPEG processing high ISO NR that the manual is talking
about and that doesn't affect the raw files no matter where it is
set. What we are talking about here is NR applied to the raw files
that can't be turned off for ISO's of 1600 and higher.

As Oleg says, it only reduces detail levels a little bit at the same
time as it reduces noise and doesn't totally destroy the
images, but raw purists would rather it wasn't there.

Regards, GordonBGood

I would love to read actual proof of what you say. Where is this
information coming from? Because it's certainly not coming from the
product manual. The blurb of ISO3200+ = High NR says absolutely
nothing about whether this is only applied to RAW or JPEG shots.

Seriously - If someone can produce a link I would love to read it.

No, the raw NR isn't in the manual and likely isn't something that Pentax would like to advertise anyway. Neither did Sony advertise that their first firmware version of the A700 absolutely clobbered the high ISO raw data files with NR. At least the K20D NR is nowhere near as strong and only will decrease high ISO resolution slightly.

The proof is in analysis of the statistical nature of the raw data in that adjacent photosites in the raw files from the K20D are more statistically related to each other than for the raw files from other cameras. Oleg produced some mathematical analysis work and presented it on a Russian PentaxClub forum, with parts of it presented here, but we didn't present it here because no one seemed to want to accept our work.

Regards, GordonBGood

Keyboard shortcuts:
FForum MMy threads