Has Pentax trumped Nikon crop sensor cameras at high ISO?

q.g._de_bakker · August 29, 2013

Laurentiu, You do get more noise. Yes, you can check signal to noise ration specs, and find that "measurements of SNR on the RAW images under equivalent shooting conditions give similar results". Same level of noise per pixel. More pixels. More noise. The linked bit of internet filling even mentions that you get more noise in in the opening line: "despite the increase in noise." What it is going on about are post-processing methods to reduce noise. Now who would have known that it goes against conventional wisdom that you can reduce the visibility of noise in post-processing... And that post-processing tricks that depend on reducing the amount of image information work best (as in: leave a still acceptable result) if you have lots of that to begin with. Thanks DxOMark! There is a very good reason why the "conventional wisdom" does not agree with how you interpret what that text says. And that reason is that, unless you recognize it for what it is, it is wrong. And what it is, is no more than a description of how you can use post-processing to reduce the visible effect of noise [etc.] A bit of misrepresentation discussions like this can do well without.

laur1 · August 29, 2013

With a Bayer sensor it becomes even trickier because you replace a red, green, or blue photosite with 2 greens, one red, one blue. You lose half the ability to capture green but you gain more ability to capture red and blue

</blockquote>

I thought a bit more about this. What I said here holds for a comparison of one large photosite with 4 smaller ones, but actually, if you extend the comparison to look at 4 large photosites (RGGB) vs 16 smaller ones, you'll find out that they capture the same light, only the smaller photosites are perhaps better capable to record fine color gradation. In practice though, I expect that there are many more aspects to light collection that make a theoretical model of how a sensor operates much more complicated.

Anyway, here is one other follow up article to the one I mentioned before - unfortunately, this is from 2008, so it doesn't take into account newer APS-C sensors like that of the Pentax K-5/Nikon D7000, so the picture of APS-C sensor development does not look as good as it looks now:

<a href="http://www.dxomark.com/index.php/Publications/DxOMark-Insights/SNR-evolution-over-time">The paradoxical evolution of sensor SNR over time</a>

Tim_Lookingbill · August 29, 2013

The very reason I initially posted that Canon 7D vs Pentax K-5 comparison was that I found it interesting that you could under expose a scene by 5 EV and still recover a decent image with very little noise and loss of detail. I mean who shoots like that. I don't, but it was nice to see it work with a Pentax but trip up another competitor.

But what that comparison also showed me was how different two cameras behave in this manner which I was under the impression all camera's were pretty much the same.

Now if the Nikon D7100 (or which ever Nikon Lannie chose) can recover the image the same way as the Pentax K-5 over the Canon then that's just even more info that says some cameras ARE the same. But I still say you can't separate the causality of this from the hardware vs software's influences beyond a shadow of a doubt nor is it useful to a photographer to do so.

laur1 · August 29, 2013

The linked bit of internet filling even mentions that you get more noise in in the opening line: "despite the increase in noise."

</blockquote>

Which opening line? I cannot find this phrase in the opening line of either dxomark article. Can you please quote with context and reference to page in article, if this is not in the "opening line"?

What it is going on about are post-processing methods to reduce noise.

</blockquote>

Where did you get that idea from? Their measurements specifically point out noise reduction, when detected, as in the case of Pentax cameras. Why would you think that the D7100 showing performance as good or better than the D7000 is due to noise reduction being applied to the D7100 RAW files?

The very reason I initially posted that Canon 7D vs Pentax K-5 comparison was that I found it interesting that you could under expose a scene by 5 EV and still recover a decent image with very little noise and loss of detail

</blockquote>

That is due to K-5's 14.1 EV DR vs Canon 7Ds 11.7 Ev DR. The Nikon D7100 has 13.7 EV DR - a bit less than the K-5, but in practice you can probably do similar tricks with it. It's the 7D sensor that is particularly weak in this area.

q.g._de_bakker · August 29, 2013

"Which opening line? I cannot find this phrase in the opening line of either dxomark article. Can you please quote with context and reference to page in article, if this is not in the "opening line"?" <h2>"Contrary to conventional wisdom, higher resolution actually compensates for noise</h2>This Insight uses specific DSLRs to demonstrate the technique for objectively comparing noise for cameras with different levels of resolution. Such comparisons conclusively show better results overall for high-resolution sensors, despite the increase in noise." Have you really read the article, Laurentiu?

laur1 · August 29, 2013

This Insight uses specific DSLRs to demonstrate the technique for objectively comparing noise for cameras with different levels of resolution. Such comparisons conclusively show better results overall for high-resolution sensors, despite the increase in noise." Have you really read the article, Laurentiu?

</blockquote>

Yes, no need to be sarcastic - I just couldn't find your snippet for some reason - I must have copied it with some extra text when I searched it in the article. They are talking about noise at pixel level, as a later paragraph shows:

Basically, the same amount of light captured by the optics will be shared by more pixels, and since each individual pixel will receive less light, the signal-to-noise ratio (SNR) will be lower. In short, you will end up taking photos with more noise.

</blockquote>

The entire article describes their testing approach where they scale down results to a common print size. But the noise reduction (SNR increase) comes from the scaling down, not from any specific noise reduction algorithm:

Because of the very high resolution of the 1Ds Mark III, the printer will downsample the image and decrease the noise, giving a clear advantage of about 3dB to the 1Ds Mark III.

</blockquote>

The SNR they calculate for this scaled down image they refer to as "normalized SNR". As you can see from the tables in the real life comparisons section (and as they explained earlier), the "normalized SNR" is higher than the pixel level SNR, unless the camera has less resolution than their reference resolution. In those examples, you can see how the SNR of the 1DsMkIII gets to be pretty close to that of the D3, although it is still inferior. But then that article was written in 2008 and since then we had several developments in sensor technologies, which is what allows the D7100 to be better now than the D7000 without any noise reduction tricks. In 2008, light would be lost in between photosites, but now this is less of an issue in designs using gapless microlenses. Back illuminated sensor designs are another improvement making photosites more efficient. Where we got now is that sensors are close to theoretical performance - i.e. they can capture most of the light falling on their surface. And that means that the main factor that determines noise is sensor size, not pixel density. Sure, there are limits around how large a pixel density you can afford, but they are no longer of much practical concern. Take the<a href="http://www.dxomark.com/index.php/Cameras/Compare-Camera-Sensors/Compare-cameras-side-by-side/(appareil1)/792%7C0/(brand)/Nikon/(appareil2)/441%7C0/(brand2)/Nikon"> D700 and the D800</a> - the latter has three times the resolution and still gets better normalized SNR - it is actually better any way you look at it - better DR, better color depth too. What you claim about the benefit of larger photosites used to be true. But it wasn't because there was a law of nature enforcing it, it was simply because implementations were inefficient. As technology changed, some things stopped being true. This is one of them. There is no noise reduction magic going under the hood. There is only the benefit of more efficient photosites and I suspect this is combined with a big benefit in demosaicing as well, because a finer pattern will be cleaner when scaled down.

q.g._de_bakker · August 30, 2013

No sarcasm, Laurentiu. The article talks about more noise (lower SNR and more, noisy, pixels) at higher resolutions, and how to 'normalize' for comparisons. Contrary to DxO's wisdom, the noise levels you get are the noise levels you get. Their 'normalization' is just a way to misrepresent things. Their answer to their question "Which is better, a low-resolution camera with less noise, or a high-resolution camera with more noise?" is sought in a way to reduce the high resolution image to a lower resolution one, and then - lo and behold! - in reducing the image information, you also lose some of the noise (as in" "The loss of resolution produces a better SNR." "We now have [...]" indeed. Duh... No need to look at the results, or at the real data before 'normalization'. "Higher resolution actually compensates for noise", if you sacrifice higher resolution. Their 'normalization' is just a reduction of that which is 'more' (resolution, noise). "To compare prints on a given format of cameras with different resolutions, [...]" they throw away what turned the higher resolution camera into a higher resolution camera. The method is one that is contrary to conventional wisdom, because it's not a very wise one.

Tim_Lookingbill · August 30, 2013

So basically you either sacrifice resolution by applying a scaling down algorithm blender to bust up, dilute and distribute the noise within the detail or retain full resolution and keep more noise.

I'll keep the resolution. Noise size at that resolution isn't going to be seen on a stochastic dithered print anyway.

q.g._de_bakker · August 30, 2013

Did i leave a tag open? As it appears, i did. Sorry!

Landrum Kelly · August 30, 2013

I'll keep the resolution. Noise size at that resolution isn't going to be seen on a stochastic dithered print anyway.

</blockquote>

Exactly. If you can't see it, what difference does it make?

On the other hand, I never know when I am going to do a radical crop and need a really clean pixel-for-pixel image. Sometimes that crop does not have to be so radical for me to begin to see the noise.

I think all of you guys have a piece of the truth on this issue.

--Lannie

laur1 · August 30, 2013

The method is one that is contrary to conventional wisdom, because it's not a very wise one.

</blockquote>

Really? So I am looking at two cameras: one that gives me 36MP and one that gives me 12MP. Let's say all else is equal (price, body features, etc). Furthermore, the 36MP camera produces 12MP images just as good as the 12MP camera (this is what the "not very wise" dxomark tests are showing me). And you say I would be wise to ignore that and insist on getting the 12MP camera because it produces ... what exact IQ benefit?

Conventional wisdom insists on recommending 12MP even though the 36MP provides the same IQ. And then other masters of conventional wisdom argue that the reason of being of APS-C is to provide more reach over FF by virtue of having smaller pixels. That is the problem with conventional wisdom - it contradicts itself, because it's not even close to being real wisdom.

The only drawback with high resolution cameras is storage space. But that is a different point that has nothing to do with IQ.

Landrum Kelly · August 30, 2013

Laurentiu, it is interesting to me that you compare a hypothetical 12-mp camera with a hypothetical 36-mp camera. I have one non-hypothetical version of each, the Nikon D3s (12 mp) and the Nikon D800E (36 mp).

Pixel for pixel, the D3s really is cleaner, but guess what: even if you down-sample the huge files from the D800E, the D3s still gives better low-light, high ISO results, in spite of the fact that its technology is older than that of the D800E.

Theory aside, those are simply the facts. The difference will not be great at low ISO, but at high ISO all those tight-packed pixels do begin to interfere increasingly (I believe) with each other. I like both cameras, but they really are not the same at high ISO. Sometimes I think that noise (interference) must increase exponentially (or at least disproportionately) as one turns up the juice (ISO). That brings on the need for strong noise reduction--and that is what we tend to see in too many high megapixel cameras.

This is why MF digital will always be superior to anything put on a 36x24mm sensor: the pixels are far apart enough that there is much less interference, i.e., noise.

Even the Pentax cameras that I started this thread with do better than many others because they have modest pixel density.

--Lannie

q.g._de_bakker · August 30, 2013

It is not very wise, Laurentiu, (and DxO say so themselves when they say it goes against conventional wisdom - they are not more clever than everyone else. Quite the contrary), because it says that two different cameras have a comparable amount of noise, if you reduce one to the level of the other. So unless you are always going to reduce what a higher resolution camera produces down to what a lower resolution camera produces, it means nothing (except that when you always reduce one to the level of the other, you then get what you then get). It would indeed be wise to use a high resolution camera for the benefits it brings (for instance those "more pixls per duck"), and not throw those benefits away (unwise to do so) to gain a disproportionally small benefit (a small "reduction" in noise). DxO's article isn't saying you should. They only say that should you do so, should you throw a large part of that thing called image quality away, you can find a ' normal' that you can use to compare one aspect of image quality (which - as should be clear - is not true: the comparison only holds if and when you do, if you get rid of some noise by downsampling a higher resolution image). Does conventional wisdom insist on recommending 12 MP over 36 MP, Laurentiu? Since when would that be? But yes: again, if that recommendation is based on the assumption (!) that both provide the same IQ"... What do they say again? Duh. Which is all too rife in this thread: the assumption that higher MP images are no better than lower MP images. It's of course very easy to argue, on that assumption, that lower MP cameras are just as good as higher MP cameras... If you first ignore or 'argue away' the benefits one thing has over one other thing, it is of course extremely easy to point out the benefits that other thing then may have over the first. But if you want to discuss how things really are, we really should get rid of such presuppositions. Misrepresentations and poorly understood other 'goes-against-conventional-wisdom' misrepresentations only help create and maintain misunderstanding.

laur1 · August 31, 2013

Pixel for pixel, the D3s really is cleaner, but guess what: even if you down-sample the huge files from the D800E, the D3s still gives better low-light, high ISO results, in spite of the fact that its technology is older than that of the D800E.

</blockquote>

Can it be because the DR (and color sensitivity to a lesser extent) of the D3s actually holds up better at high ISO as dxomark shows? This is what it looks like to me from the dpreview test shots too. And tell me this: how do you find the comparison at low ISO? dxomark indicates a strong advantage for the D800E and not only because it offers ISO 100 which D3s is lacking. That advantage seems to hold up to ISO 800 and only beyond that does the D3s start to show its advantage. Also, what I pointed out via that dxomark article is a trend. One case that goes against the trend won't negate it. My point was simply that people shouldn't think that just because a camera has more pixels it will deliver worse performance. That doesn't mean that I believe the opposite is always true (i.e. if a camera has more pixels, it will deliver better performace). In the case of the D800 vs the D3s I'd still go for the D800 because I wouldn't be willing to sacrifice low ISO performance for better performance at ISO settings at which I shoot <5% of the time.

So unless you are always going to reduce what a higher resolution camera produces down to what a lower resolution camera produces, it means nothing

</blockquote>

Of course it means a lot. If you print or display on a monitor, you will either have to scale down the large resolution image or you will have to blow out the small resolution image - you cannot avoid doing one or the other.

q.g._de_bakker · August 31, 2013

"Of course it means a lot. If you print or display on a monitor, you will either have to scale down the large resolution image or you will have to blow out the small resolution image - you cannot avoid doing one or the other." You can. With great ease. Print to the size the amount of pixels the camera produces allows. But that's neither here nor there. More important (and i'll ask you, Laurentiu, instead of explaining myself) is what it would then mean.

Landrum Kelly · August 31, 2013

"More important (and i'll ask you, Laurentiu, instead of explaining myself) is what it would then mean."

</blockquote>

What does that mean?

Q.G., at times I think that you are a master of obscurantism.

Perhaps, after several weeks, this thread has pretty much run its course.

Thanks to everyone for all the comments. I would like to try one of the low-light Pentax cameras, but I simply cannot afford that experiment right now.

--Lannie

q.g._de_bakker · September 1, 2013

Landrum? Laurentiu says "it means a lot". I ask what that "a lot" would be. Obscure? The only obscure thing is what meaning would be hidden away in that DxO misrepresentation. I agree that this thread has run its course. Some cameras are good. some are better. Some are not so good. Some considerations offered in this thread, for instance those concerning pixels per duck and conventional wisdom, were also not so good. You have to compare.

Landrum Kelly · September 1, 2013

Holy cow! It's September.

--Lannie

Landrum Kelly · September 1, 2013

Some considerations offered in this thread, for instance those concerning pixels per duck and conventional wisdom, were also not so good.

</blockquote>

"Pixels per duck"--yes, the preferred highly technical industry term. Thank you for that one, Q.G.

--Lannie

q.g._de_bakker · September 1, 2013

Render unto Keith the things that are Keith's, Landrum.

laur1 · September 2, 2013

Print to the size the amount of pixels the camera produces allows.

</blockquote>

Q.G., you are making me curious: where in the world have you found people that print like that? Is this a cultural thing somewhere?

More important (and i'll ask you, Laurentiu, instead of explaining myself) is what it would then mean.

</blockquote>

Already told you:

If you print or display on a monitor, you will either have to scale down the large resolution image or you will have to blow out the small resolution image - you cannot avoid doing one or the other.

</blockquote>

At this point, if you don't get it, it's because you don't want to. All the arguments have been presented and there is no point in me repeating them.

q.g._de_bakker · September 2, 2013

Is it a cultural thing, Laurentiu, not to spend loads of money on megapixel cameras when you're only going to use as much of it as a cheaper thingy would have given you to play with? Well, maybe it is. Which brings us straight to your "Already told you": if you are going to buy a camera to display on a monitor, why on earth are you discussing the things you are discussing here? Laurentiu, there is only one good reason to spend money on high resolution cameras. And that single reason is that you need the high resolution. And if you do, that DxO nonsense is easily recognised for what it is: nonsense. Noone of sane mind is going to throw away expensively bought pixels because DxO stumbled upon a method to reduce noise a bit. And as mentioned before: noone too will believe it particularly interesting to hear that if you reduce one thing to the level of another thing, the result is that both are at a comparable level. Let alone think that significant. So again, Laurentiu, what hidden meaning, contrary to conventional wisdom, have you detected? Do share!

laur1 · September 2, 2013

Is it a cultural thing, Laurentiu, not to spend loads of money on megapixel cameras when you're only going to use as much of it as a cheaper thingy would have given you to play with?

</blockquote>

When all else fails, bring up the cost. :)

Cost doesn't matter Q.G. - we got that out of the way earlier and you still weren't happy:

Let's say all else is equal (price, body features, etc).

</blockquote>

So why wouldn't you want a camera with more resolution and same or better IQ for a given print size, for the same price? I provided the answer for that too:

The only drawback with high resolution cameras is storage space. But that is a different point that has nothing to do with IQ.

</blockquote>

See, going in circles now.

Which brings us straight to your "Already told you": if you are going to buy a camera to display on a monitor, why on earth are you discussing the things you are discussing here?

</blockquote>

Most images today are seen on monitors, not exchanged as prints. Give it a decade or two and you won't even have prints anymore - only thin, light displays.

And as mentioned before: noone too will believe it particularly interesting to hear that if you reduce one thing to the level of another thing, the result is that both are at a comparable level. Let alone think that significant.

</blockquote>

No one? Really? Whatever.

Landrum Kelly · September 2, 2013

Most images today are seen on monitors, not exchanged as prints. Give it a decade or two and you won't even have prints anymore - only thin, light displays.

</blockquote>

I suspect that there will still be a market for framed prints, if only as cultural artifacts. For some reason, persons like tangible things that they can call their own.

--Lannie

Landrum Kelly · September 2, 2013

No one of sane mind is going to throw away expensively bought pixels because DxO stumbled upon a method to reduce noise a bit.

</blockquote>

Q.G., DxO discovered nothing with regard to how to reduce noise. The point is that, even if one buys a 36-mp camera for its resolution, occasions are going to arise when one is going to need the camera in low light. At such times, down-sampling files will substantially reduce visible noise. Simply because one can print huge files does mean that one is obligated always to do so, especially if other considerations are more compelling.

I shoot at two extremes of full-frame right now: the D800E and the D3s. The D3s is better in low light. Yet, I will not always be carrying both, and, if I am only carrying only the D800E, it is good to know that it can do quite well in low light by down-sampling. No, it will not beat the D3s, but it does quite well.

If I had to sell one right now, I would sell the D3s. The D800E can handle most situations quite well, with the exception of high speed, fast action shots. The D3s is never going to give me more megapixels, which I often need for landscapes. It is a great camera for those with specific needs. For many of us, it is not optimal, although it is indeed a dream to shoot, and the files are always manageable.

What is so hard to understand about down-sampling when the situation requires it, Q.G.? A lot of this thread has been devoted to that issue, and you have yet to make clear what the core of your argument is, in spite of your many words.

--Lannie

Sign In

Has Pentax trumped Nikon crop sensor cameras at high ISO?

Recommended Posts

Link to comment

Share on other sites

Top Posters In This Topic

Popular Days

Top Posters In This Topic

Popular Days

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Create an account or sign in to comment

Create an account

Sign in