[slartibartfast01]

Is it? Black detection and subtraction is about getting rid of the noise that results from interference, as measured by a black frame exposure. Gain adjustment is about noise being increased by the gain to extrapolate the exposure (like pushing underexposed film in the old days). Defect correction would deal with noise resulting from defect pixels that don't respond to the gain. And the noise reduction unit would take all that to effectively reduce noise.

Could be just about detecting different types of noise and feeding that into the noise reduction unit? All of the types mentioned seem to me in-camera generated types of noise, and not noise from fluctuations in the light. Isn't it removal of the latter that can cause detail loss?


Presented like this, the decomposition into steps actually reassures me that the noise reduction done by the accelerator is perfectly OK. And all noise from fluctuations in light will still be present after the accelerator has done its work, and as such can be done by the JPG engine in the PRIME or with perfectly configurable post-processing tools.

It does sound like dark frame subtraction but surely they can't be doing that for every shot. You wouldn't notice at fast shutter speeds but it wouldn't be acceptable at slower speeds.

[Class A]

"Taking sides"?

I think when you write (emphasis is mine)

"By now most of us understand the arguments even if we've not ever seen a detrimental effect from it in our own photos, or anyone else's in my case."

you are suggesting that arguments against the "accelerator unit" are unfounded.

I feel that you could have made your moderation input without injecting which side you agree with. Of course you can express your view any time you like when not posting in a moderator role, I'm just saying it would make sense to separate moderation activities from sharing personal views.

The discussion became far more than "is the accelerator unit optional".

Correct, but I think posters like myself should have the right to respond to incorrect statements or misinformation.
I do not see why I should let misleading posts stand unopposed just in order to avoid a debate about the "accelerator unit" (which I did not invite in this thread).
In short, I don't think there are any grounds for criticising my part in this discussion.

That question was answered: None of us know 100% yet.

Thanks to Kobie, we know the real answer (that the NR settings don't affect RAW files), but that answer only came in very recently.

You weren't singled out, the moderator note is for everyone participating in this thread.

I was not singled out but I do not know why I should be included.

What is wrong about

• dispelling the myth that "Pentax engineers" said it would be impossible to turn the "accelerator unit" off?
• pointing out when posters make incorrect claims?
• making verifiably correct technical arguments about the nature of the "accelerator unit"?
• defending one's posts against invalid criticism?

Remember that all this happened in a thread I created about the "accelerator unit" and my views on the device should come as a surprise to nobody. If some posters still invite themselves to make posts that border on being disparaging and slanderous then I believe the moderation focus should be on their discussion style.

I just do not understand the notion of apparently wanting to suppress a topic just because some posters cannot contribute in a civilised manner.

The longer the very old "it does this, no it doesn't, yes it does, no it doesn't" goes on between three or four members the higher-pitched the voices have become which serves no one well.

I submit that my voice has never been a "high-pitched" one.
I further submit that this is not just a case "X says A" and "Y says B" and there is no way of telling who is right. Some claims can and have been refuted and some remaining ideas, e.g., that denoising techniques exist that leave detail perfectly intact are just not tenable.

Do you feel there's something new and still pertinent you wanted to say that hasn't been said?

I think a thread in which information is made available that dispels the myth of Pentax engineers having made a strong claim about the "accelerator unit" is useful. Clearly, there were still a number of posters who held incorrect views about this. I also note that some of my posts have been appreciated by some. Finally, I've used a new example to illustrate the technical facts underlying the discussion.

[Class A]

As promised...
25,600 ISO, DNG, NR Off, Low, Med, High. Exact same scene for all images.
Really no difference

Thanks a lot.

Too bad that there is no effect on the RAW data because it would have been a win if every side could have chosen their preferred option. I am very concerned that DPReview will use the mandatory RAW data manipulation to deny the K-3 III the recognition it deserves.

[Kunzite]

Thanks to Kobie, we know the real answer (that the NR settings don't affect RAW files), but that answer only came in very recently.

Check post #5

[Class A]

So NR is only applicable to JPEG's which I was not aware of.

Yes, the additional the additional "High-ISO NR" has now been shown to only apply to JPEGs.

There is still denoising though of RAW data (performed by the "accelerator unit II") because we'd otherwise see RAW files that are more comparable to those of the K-1 II. The K-3 III uses a BSI-sensor which has a bit better noise performance than a standard sensor but that advantage is not big enough to achieve the gains over the K-1 II without applying in-camera image processing (which Pentax openly admit to).

[Class A]

The signal processing is more than just NR, but that seems to be the only thing people focus on.

Speaking for myself only, I "focus" on processing that is destructive.

It is irrelevant to me whether the "accelerator unit" performs other functions or applies processing that is irreversible when criticising the destructive nature of one of its functions.
I am not against the use of the "accelerator unit" in a camera. I'm just saying for those who know why they don't want denoising applied to their RAW files, the option to bypass the processing should be there (contrary to what have been claimed by some, the bypassing is clearly possible (cf. the patent description)).

...but at the same time leaves enough of the "roughness" to ensure the detail isn't lost.

I'm genuinely glad for you that you judge the processing to be moderate enough so that you don't have any issues with loss of detail.

It is, however, technically undeniably true that any form of post-denoising, will inevitably remove some of the original signal as well.
Even if it is below the threshold of visibility for a single image (which it was not for the K-1 II, in comparison to the K-1 one could identify some slight smearing of details at times which a pentaxforums.com article reported on), then it can make a difference in applications like deep sky astrophotography where the residual signal in the noise builds up to visible detail again after many images are combined together.

Accurate colors, detail retention and a cleaner image at ISO levels we wouldn't dare venture into a few years ago. I'll take it!

I agree that the option to get the in-camera processing is great and I might use it myself from time to time.

All I'm saying is that there are applications in which the in-camera processing is not helpful, and this is only repeating what Pentax themselves said about the "accelerator unit".

[Class A]

Presented like this, the decomposition into steps actually reassures me that the noise reduction done by the accelerator is perfectly OK.

There are certainly ways to achieve noise reduction that are completely non-contentious.

Unfortunately, in the case of the K-1 II that is not the case. We'll have to await measurements for the K-3 III but I'd be very surprised if those measurements would not reveal evidence of "smoothing" gain.

The problem with the K-1 II's image processing at ISO 640 and higher is that it suppresses some detail as a by-product of reducing noise. This reduction of detail can be numerically measured and the technical approach is to evaluate the attenuation of high spatial frequencies in a 2D Fourier analysis of an image that contains pure noise. Non-contentious noise reduction techniques like black-frame subtraction do not affect the frequency spectrum of such 2D Fourier analysis. Problematic denoising techniques, however, do.

In many cases and for many people, the denoising-caused smoothing will not have any detrimental effects. For some people/applications, it would be worth having the option to disable this part of the processing.

[ThorSanchez]

Speaking for myself only, I "focus" on processing that is destructive.

It is irrelevant to me whether the "accelerator unit" performs other functions or applies processing that is irreversible when criticising the destructive nature of one of its functions.

This is why it's hard to take your arguments too seriously. You go on and on for pages and pages across multiple threads about how the accelerator's actions are "destructive". It would be natural to conclude from your statements that you could take most any reasonably high ISO file at reasonable viewing conditions and see that there's a large (or at least noticeably) negative effect. But that's clearly not the case. You have to take specific cases, magnify to 100% or 200% and know what you're looking for to discern any blurring of details at all.

You're basically telling us that your car has a 1 cm scratch on the bumper after washing and waxing it, and even though the overall appearance of the car is tremendous you now consider the paint destroyed and that everyone should always be very, very, very skeptical of washing and waxing one's car.

[Kunzite]

We'll have to await measurements for the K-3 III but I'd be very surprised if those measurements would not reveal evidence of "smoothing" gain.

The problem with the K-1 II's image processing at ISO 640 and higher is that it suppresses some detail as a by-product of reducing noise. This reduction of detail can be numerically measured and the technical approach is to evaluate the attenuation of high spatial frequencies in a 2D Fourier analysis of an image that contains pure noise.

Reduction of detail cannot be measured on images containing pure noise.
There is no measurement on the K1ii's detail loss. There will be no measurement on the K3iii's detail loss.

[Class A]

This is why it's hard to take your arguments too seriously.

First, your criticism is based on reading something into my posts that they don't contain.

Second, even if I exaggerated (which I don't) then my technical arguments stand on their own. They either make sense or they don't regardless of whether I'm exaggerating on one matter.

It would be natural to conclude from your statements that you could take most any reasonably high ISO file at reasonable viewing conditions and see that there's a large (or at least noticeably) negative effect.

I disagree that this is a "natural conclusion". I never wrote there is a "large" negative effect.
For me personally, the effect is noticeable, but I always emphasised that I'm happy for anyone who is not as detail obsessed.

You're basically telling us that your car has a 1 cm scratch on the bumper after washing and waxing it, and even though the overall appearance of the car is tremendous you now consider the paint destroyed and that everyone should always be very, very, very skeptical of washing and waxing one's car.

You are putting words into my mouth. I never said anything like that.

You are clearing reading claims into my statements which they objectively don't make.
Please revisit my posts and you'll find the opposite of "everyone" and you will not find me blowing things out of proportion.

I only said the equivalent of "after washing your car may have a 1cm scratch on the bumper" and left if for everyone to decide whether that's a problem for them or not.
I know that DPReview easily blow things out of proportion when it comes to evaluating Pentax cameras which is why I have expressed concern about the negative effect mandatory RAW denoising could have on Pentax sales. This does not equate to ringing the alarm bells for reasonable people.

Please keep your criticism factual and fair.

[Class A]

Reduction of detail cannot be measured on images containing pure noise.

You keep repeating this untenable claim.

Please educate yourself on how denoising works and what the inevitable consequences are.
Denoising is a filtering process (low pass filter) which cannot reliably distinguish between random noise and real image data.

As a slightly absurd but simple example, imagine you are taking an image of the screen of an analogue TV set which does not receive a TV signal. The scene to be captured then consists of random noise and a camera with an active "accelerator unit" now won't be able to faithfully capture that scene because it will mistake some of the real scene with noise and smooth some of the scene details.

You may now argue that there is no point in faithfully capturing random noise in a scene but the next problem is that there is no way to reliable tell shot noise from real scene detail. Some advanced processing may achieve protection of detail under certain conditions but from an information theoretic perspective it is simply impossible to get this right in every case.

Finally, some signal (i.e., image data correlating to the scene) lives under the noise threshold. If it is removed by removing the noise, it cannot be used to recreate a better representation of the scene anymore. Taking multiple exposures of the same scene and combining them to a better image (this is part of the reason why PixelShift produces cleaner images) only works because the noise in the individual images is not meaningless.

None of the above is under debate. Try pointing us to any research results that claim that denoising (post-capture noise reduction that causes smoothing in uncorrelated noise) does not affect signal. You will not succeed.

[Andrea K]

Sometimes I wonder if you look at the pictures with the eyes or with a microscope. Really, the Accelerator Unit controversy is for techno geeks (as are at dpreview) not for photographers

[gatorguy]

Final moderator note: The thread is back to where it was when the first note was issued, and that's not a good place. The pro and anti-accelerator arguments are well known and have been discussed ad-nauseum over the past three pages with no sign of resolution. Some members are now just digging in their heels. If you have something to say that has not already been said, several times in some cases, please do so before the thread outlives any usefulness.

[Andrea K]

Sorry

[slartibartfast01]

FWIW here is an example of astrophotography and image stacking using a K1 ii.
https://www.pentaxuser.com/forum/topic/m42-61693