Let's finally resolve the D800 downsampling / NR questions

If with high MP's the hi freq noise is best handled with good NR software (it works even after upsamlpling according to Dominique),
and downsampling does nothing to mid and low frequency noise,
than what's the gain in downsampling other than getting a smaller file size ?
Somewhere the image will be downsampled anyway unless you happen to be printing/viewing at its native size, so we might as well be the one who does it and control the process.
 
If with high MP's the hi freq noise is best handled with good NR software (it works even after upsamlpling according to Dominique),
and downsampling does nothing to mid and low frequency noise,
than what's the gain in downsampling other than getting a smaller file size ?
Not much. But those were the parameters of the discussion.
Wait a moment.

Downsampling is something that you do routinely to (i) make web sized images, (ii) make prints smaller than a certain size. That one should be able to achieve noise reduction benefits in these commonplaces of photography is good news. That you should be able to aggregate pixels together for purposes of signal optimization is a new consideration to many.

This latter consideration involves a cornerstone of your own teachings -- the notion that for a wide range of pixel sizes, when comparing sensors of equal sizes and with comparable technology, that overall light-gathering ability and noise are more or less equivalent per unit area of the sensor. This figures heavily into decisions about the tradeoff between, say, a 36MP D800 Exmor-Nikon sensor, and a 16MP D4 sensor in a practical way.

To wit: Can the D800 sensor be used to produce high ISO images on a par with the D4? The answer appears to be approximately yes. But to understand that, most people still need to grasp the idea that when photons are scarce, you can aggregate pixels without any in-principle penalty in order to optimize signal. In other words, the D800 makes about as good a 16MP camera as the D4 does (modulo considerations about correlated noise, the perception of detail, and work flow).
 
In theory the downsaampling should aggregate out the random high frequency noise by downsampling but leave the detail - in that detail is likely to have neighbouring pixels replicating data and so the downsampling program will likely hold more of the repetitive high frequency detail and remove some high frequency random info
...downsampling leads to loss of true detail as much as it drops the noise.

However, we humans may perceive this quite differently. We doe not see pixels (unless we start pixel peeping....) but we see patterns and regularities (color and or shapes). Noise, particularly extensive noise, simply distroys these patterns.

So, if the noise is gone to some extent by downsampling - the hi freq noise, the rest is 'unaltered'- and we still see the patterns although we also lost some of details within these patterns, we may still consider it as an improvement.
 
And with increasing MP's your conclusion becomes increasingly important.

But so far, my conclusion is that it wil not help you much in terms of noise reduction unless you accept loss of detail as well. And, that with high MP sensors NR software does the process more efficiently than downsampling.
Right ? Or wrong?
 
And with increasing MP's your conclusion becomes increasingly important.

But so far, my conclusion is that it wil not help you much in terms of noise reduction unless you accept loss of detail as well. And, that with high MP sensors NR software does the process more efficiently than downsampling.
Right ? Or wrong?
Right in some cases to be certain. But there are additional considerations.

Sometimes the outright scarcity of photons suggests that aggregating your signal into a smaller number of pixels will give you a more optimal output.

You could get a D4, in which the photons are recorded in 16M relatively large containers, and pixels are all practically usable at 100%. But I think I would rather record 36M pixels archivally and decide later how I'd like to treat them. Future advances in DSP techniques may yield better image treatments, if the last five years are any guide to the future.
 
Downsampling is something that you do routinely to (i) make web sized images, (ii) make prints smaller than a certain size. That one should be able to achieve noise reduction benefits in these commonplaces of photography is good news.
Just to be clear. You mean that when you routinely down sample you magically get D4 performance? (how does I don't think so sound).

Cheers,

-Yamo-
 
Downsampling is something that you do routinely to (i) make web sized images, (ii) make prints smaller than a certain size. That one should be able to achieve noise reduction benefits in these commonplaces of photography is good news.
Just to be clear. You mean that when you routinely down sample you magically get D4 performance? (how does I don't think so sound).
I'd say only with the D800 does this work, and not by magic. But you can get D3 performance out of a D3x, when downsampled to 12MP -- not D3s performance, but D3 performance.
 
If with high MP's the hi freq noise is best handled with good NR software (it works even after upsamlpling according to Dominique),
and downsampling does nothing to mid and low frequency noise,
than what's the gain in downsampling other than getting a smaller file size ?
Not much. But those were the parameters of the discussion.
Wait a moment.

Downsampling is something that you do routinely to (i) make web sized images, (ii) make prints smaller than a certain size. That one should be able to achieve noise reduction benefits in these commonplaces of photography is good news. That you should be able to aggregate pixels together for purposes of signal optimization is a new consideration to many.

This latter consideration involves a cornerstone of your own teachings -- the notion that for a wide range of pixel sizes, when comparing sensors of equal sizes and with comparable technology, that overall light-gathering ability and noise are more or less equivalent per unit area of the sensor. This figures heavily into decisions about the tradeoff between, say, a 36MP D800 Exmor-Nikon sensor, and a 16MP D4 sensor in a practical way.

To wit: Can the D800 sensor be used to produce high ISO images on a par with the D4? The answer appears to be approximately yes. But to understand that, most people still need to grasp the idea that when photons are scarce, you can aggregate pixels without any in-principle penalty in order to optimize signal. In other words, the D800 makes about as good a 16MP camera as the D4 does (modulo considerations about correlated noise, the perception of detail, and work flow).
Perhaps I was too terse. I understood lock's point to be that, if high frequency noise is best handled by dedicated NR tools, and downsampling does nothing to mid and low frequency noise, then what added benefit is there to downsampling; and indeed there is little. If however, one is not going to use NR tools, then downsampling will have the effect of reducing high frequency noise (and signal, of course).

As to the rest, it's early days yet, but I expect that the D800 will be a very good low light performer. One thing that would be interesting to look into is how high frequency noise interacts with demosaic. Just anecdotally, the impulsive character of some read noise can wreak havoc on some interpolation methods, IME.

--
emil
--



http://theory.uchicago.edu/~ejm/pix/20d/
 
If with high MP's the hi freq noise is best handled with good NR software (it works even after upsamlpling according to Dominique),
and downsampling does nothing to mid and low frequency noise,
than what's the gain in downsampling other than getting a smaller file size ?
Very few people would suggest downsampling as a preferred method to reduce noise. We talk about the effect of downsampling on noise because we often have to downsample anyway, eg, when viewing an image on a monitor, or when printing at small to moderate sizes or when sending images via e-mail or posting it on the web.

Thus, if we downsample anyway (for a large proportion of out images), it makes little sense to talk about the noise before downsampling (that applies to basically all images viewed via the web or on mobile devices).
 
That's exactly what I meant.

The big question is: which one does the better job ? NR with high Mp kept in tact or downsampling and ending up with a lower resolution ? You probalby know how I would answer this question.

lock
 
If with high MP's the hi freq noise is best handled with good NR software (it works even after upsamlpling according to Dominique),
and downsampling does nothing to mid and low frequency noise,
than what's the gain in downsampling other than getting a smaller file size ?
Not much. But those were the parameters of the discussion.
Wait a moment.
To wit: Can the D800 sensor be used to produce high ISO images on a par with the D4? The answer appears to be approximately yes. But to understand that, most people still need to grasp the idea that when photons are scarce, you can aggregate pixels without any in-principle penalty in order to optimize signal. In other words, the D800 makes about as good a 16MP camera as the D4 does (modulo considerations about correlated noise, the perception of detail, and work flow).
Luke,

You are offering up a very interesting conjecture. Whether this conjecture is correct can only be ascertained by practice.

On a related note, on LL MF, just everybody has been saying "Oh, give us fewer better pixels". If these guys (including me) had a magical downsizing trick that really really worked, we'd know it. I have a 39 MP camera that is long out of warranty, as you know. And my D3x, below a certain amount of ambient light, it kind of turns into mush whatever I do.

Edmund
--
Ouch, my name is mistyped - my name is Edmund Ronald
 
At least not, if there is no specific reason to do so.

The screen is something different. Yes, we downsample. Until we get higher res screens and/or bigger ones, we will downsample.

But I do not take pictures just to display them on my screen. I'm too old-fashioned for that :)
 
If these guys (including me) had a magical downsizing trick that really really worked, we'd know it. I have a 39 MP camera that is long out of warranty, as you know.
It only works if the 36 (or 39) MP sensor is good enough in the first place. Your sensor probably has a quantum efficiency of under 20% and a read noise of around 15 electrons per pixel. Whatever you do you won't make it match a modern CMOS sensor with a QE of 50%+ and a read noise of about 2 electrons.
--
Bob
 
At least not, if there is no specific reason to do so.
I of course meant when printing at a size which at the printer's native resolution is below that of your camera, ie, as I said before, we downsample when we have to downsample.
 
Must have missed the point. You are of course correct. There's no avoiding the downsampling in that case either.
 
Luke,

You are offering up a very interesting conjecture. Whether this conjecture is correct can only be ascertained by practice.
Judging from samples, it looks very promising. If I can take an ISO25600 image from a D800 and squeeze another stop out of it, to the point where it looks at least as good as my D3s, that's very promising. This much I can do now.

Though I don't know why for the life of me all the people who post samples are such lousy photographers. I mean, jeez, if you've got a D800 in your hand, take a picture of something other than the contents of your pocket. Big world out there.
On a related note, on LL MF, just everybody has been saying "Oh, give us fewer better pixels". If these guys (including me) had a magical downsizing trick that really really worked, we'd know it. I have a 39 MP camera that is long out of warranty, as you know. And my D3x, below a certain amount of ambient light, it kind of turns into mush whatever I do.
If it weren't for the Exmor, 3rd generation all the more, I'm not sure any of the "in-principle" stuff would have come into the realm of the practical. I don't know how this works out in current CCD implementations. I think we need a medium format Exmor chip.
 

Keyboard shortcuts

Back
Top