gollywop
Veteran Member
Ah, yes – alliteration. and just a touch of near assonance, which, in the case, may lend a touch of onomatopoeia.
thank you.
--
gollywop
-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
How does "total light" change SNR? II
- Thread starter Great Bustard
- Start date
gollywop
Boy, there are a lot of layers to this onion. Thanks for making me think.
First, a quibble. If you consider vacuum tube amplifiers another case, Walter Schottky seemed to have considered shot noise "noise" when he discovered the effect. BTW, yes, this same Schottky that gave his name to the Schottky transistor, and he lived to see Texas Instruments and others commercialize it in a big way with the 74S and 74LS TTL series.
Second, I admire the purity of your argument, but I think noise as most photographers use the term means unwanted data in the photograph. Yes, I know that just kicks the can down the road. What's unwanted; what's wanted, etc. In many situations, photon noise is unwanted, and photographers go to physically bigger sensors, turn down the ISO knob and increase the exposure, etc, to reduce it. Oops, actually those things increase the photon noise, but they increase the wanted information faster, so the wanted to unwanted ratio (to try to avoid using noise for photon noise -- ugly, isn't it?) gets larger.
Unless somebody can come up with a great set of words to replace noise in photon noise and in SNR, I think this is a wormhole to an angels/pin discussion.
Jim
As an aside, in the past, when I've needed a random or pseudo-random signal for some device or system I'm designing, I've used shot noise or fed-back shift registers to get it. I've always called the line(s) bearing the signal by some designator with noise in its name. In this patent , I actually called it a noise signal, which, in your way of looking at the world, is an oxymoron. If I'm to get with your program, since the line carries precisely the information I want it to carry, I can't call it noise. I guess I could have called it something like stochastic signal. Kinda makes my head hurt.
Great Bustard
Veteran Member
- Messages
- 45,961
- Solutions
- 17
- Reaction score
- 34,046
It doesn't matter how many pixels that greater amount of light is distributed over. More pixels simply greater accuracy in sampling the signal. More light means the closer that sample is to the mean signal.
What's unfortunate is that some take their lack of understanding as an example of being "misleading".
Great Bustard
Veteran Member
- Messages
- 45,961
- Solutions
- 17
- Reaction score
- 34,046
I was pretty sure he was Eric Fossum posting with another nick. :-D
gollywop
Veteran Member
Ha! Well, Eric is almost clever enough to carry that one off. But it's hard for even the cleverest person to play that stupid. So don't worry Eric; you're off the hook. ;-)
Meanwhile, Joe, you lose; you were lied to about being put on the ignore list.
I have never placed anyone on an ignore list, but I sure have a fairly well populated ignore-ance list.
--
gollywop
D
dtmateojr
Guest
Thank you for the extensive background check.I'm beginning to realize that this dtmateojr guy (Demosthenes Mateo Jr.) is simply in seventh heaven playing in the sandboxes he creates here on dpr. After observing his behavior in a couple of recent threads, I became curious to figure out what might be motivating it, and I believe that it that he is just plainly overjoyed at having finally found a way to get people to talk to him. He needs attention at any cost.
It is of interest to note, for example, that his twitter page, which began in May 2011, has 69 tweets and absolutely zero replies over two years. His last tweet was July 6, 2013, shortly after his joining dpr on June 13, 2013. In this new venue he has clearly found a way to attract attention, and as long as he continues to get it, I am convinced he will just go on and on with his current purposely inane and obtuse behavior.
Bear this in mind when you consider responding to his posts. He isn't going to change. He has no reason to. He has what he wants/needs, and you will just be feeding its continuation.
You will also note that the comments made to his famously fatuous post, to which he frequently refers, are best summed up with reilly diefenbach's lead-off comment: Possibly the silliest post having to do with photography anyone has ever made.
He is apparently also (are you ready for this?) a RedHat Certified Engineer, an accomplishment for which he is inordinately proud. You can glean another dimension of his personality from his response to the question as to what being a RHCE means to him. You will note that all the others on the page give responses that actually answer that question. dtmateojr's response, however, is nothing but a self-serving brag session that indicates that the only meaning for him of being a RHCE is in its providing an exam or two on which he could get perfect scores.
Play in his sandbox if you will, but recognize what's buried there.
--
gollywop
Correction, I am.a Red Hat Certified Architect, Security Specialist and Virtualization Administrator as well
So yeah, I eat Linux when I'm not pretending to be a photographer. Here's my gallery http://flickr.com/dtmateojr if you want to see my other personality
May I see your photos?
Last edited:
I'm thinking of taking Junior off my list, with a new credo:I have never placed anyone on an ignore list, but I sure have a fairly well populated ignore-ance list.
Read but do not Feed
Sorta Kinda like at the Zoo . . .
--
Cheers,
Ted
Last edited:
What is being measured is the number of photons per unit area.
The error is proportional to the square root of the number. So as the photons become more numerous, the number output from a sensel becomes more certain, and the random variation between the outputs of neighbouring sensels becomes less.
See for example the Wikipedia entry on "Shot Noise".
More pixels in the same area will give less accuracy in the measurement from each each pixel, because each receives fewer photons, and therefore there is more random variation in the measurements across an array of pixels. That is, more noise, most visible in smooth areas such as sky.
Last edited:
Yes.
I think much confusion comes from talking about the signal-to-noise ratio of a single pixel in one exposure, and in the next breath about spatial noise across the image.
It is much better to talk about the uncertainty in the estimate made by one pixel in one exposure, as you say.
There is a difference here between a sensel in a camera and a microphone. Both are single devices, but the mic gives a sequence of numbers (in time) which can be called a signal. The sensel gives one number, which is better called a measurement or estimate.
bobn2
Veteran Member
The question here is what you mean by 'accuracy' in the measurement from each pixel. First you have to decide what it is that you think the pixel is measuring - if it is the number of photons, then theres not much reason to think that small pixels are less accurate at counting photons than big pixels. I suppose that you could be saying that small pixels have proportionately more read noise - but in fact the technology improvements that give us smaller pixels tend to give us smaller read noise too, so for instance the 12MP sensor in the D300 is producing about five electrons per pixel read noise, the 24MP sensor in the D5300 about two and a half, so the 'accuracy' per pixel is preserved. In any case, the read noise is a very small contributor to the overall visible noise in an image, except in the deep shadows and high ISOs - so this concept of 'less accuracy' doesn't begin to account for the noise differences observed either between same size sensors with different sized pixels (where there tends to be almost no difference, save the extra read noise, for like technology sensors) or different size sensor with same size pixels (where the performance is not, as the pixel size proponents would have it, the same).
--
Bob
Last edited:
gollywop
Veteran Member
It wasn't for your benefit. Presumably you already knew (and know) the pathos.
Oh wow! Now I'm really impressed. But it appears Linux is not very nourishing.Correction, I am.a Red Hat Certified Architect, Security Specialist and Virtualization Administrator as well
I wonder if you realize how many people you're dealing with here who have serious credentials.
I've seen 'em. Some aren't bad at all. You should stick to taking pictures.Here's my gallery http://flickr.com/dtmateojr if you want to see my other personality
Some are up in my dpr gallery, but that's mainly supportive material. On-line galleries just aren't my cup of tea. I just posted a few in another thread. And, had you been around over the years, which, thank goodness you've not been, you'd have seen a good deal more here and there. You can probably still find them if you'd care to look.
But to really see my photos, you'd have to come to my home – which you're not going to do.
--
gollywop
Last edited:
During my time, I have specified and purchased lots of Instrumentation. In that field, 'accuracy' is an important metric. It is normally expressed as a percentage or, occasionally, as actual units. Sometimes bits are tacked on to account for effects of temperature and such.
It should be so simple to define 'accuracy' for a photometric sensor, especially in this knowledgeable Forum although I, for one, will not attempt it. Enough confusion as it is ;-)
It's just surprising the amount of discussion that has been generated on that one single factor.
Great Bustard
Veteran Member
- Messages
- 45,961
- Solutions
- 17
- Reaction score
- 34,046
More pixels will sample the scene more accurately, except for when the light is so low that the read noise dwarfs the photon noise, as larger pixels tend to have less read noise per proportion of the photo, which, of course, does not happen for photos involving the sky except for night photos.
Thus, in the event we are photographing featureless scenes, then larger pixels will be less noisy (due to the lower read noise per proportion of the photo) and no less accurate because the scene itself has low resolution. However, applying noise filtering to the photo with the smaller pixels will take care of that lickety-split which still preserving more detail in the portions of the photo that do have detail.
The lower the light level, the greater the shot noise.
They will be more accurate, or rather, show less variation.
The view that software processing can fix all problems is, I think, optimistic. It would be better to minimize the problems by design.
I thinking he was referring to more pixels in the same area and by implication smaller pixels having, on average, more shot noise presumably expressed as a ratio. Unless he's lying, of course . .
I only posted really to congratulated you on your choice of the phrase 'lickety-split' which we have both used today, myself here .
W
wchutt
Guest
However you decide to define what the sensor measures, it arises from a state of nature.
I can not wrap my mind around rejecting the concept that the total light that exits a lens was not created by nature. That light has energy. The energy has a true, single value. The energy is a state of nature.
At any rate, the concept of a state-of-nature originates from the work of Edwin T. Jaynes. Jaynes was the Wayman Crow Distinguished Professor of Physics at Washington University. I wonder how his "this weasly way of appearing different " ever made it to press in peer-reviewed scientific journals?
– There is no substitute for signal-to-noise in the raw data
– Signal-to-noise can not be improved post facto
– Given a model, there are optimal methods to estimate the parameters that relate the data to the model
– There are no miracles
I can not wrap my mind around rejecting the concept that the total light that exits a lens was not created by nature. That light has energy. The energy has a true, single value. The energy is a state of nature.
At any rate, the concept of a state-of-nature originates from the work of Edwin T. Jaynes. Jaynes was the Wayman Crow Distinguished Professor of Physics at Washington University. I wonder how his "this weasly way of appearing different " ever made it to press in peer-reviewed scientific journals?
– There is no substitute for signal-to-noise in the raw data
– Signal-to-noise can not be improved post facto
– Given a model, there are optimal methods to estimate the parameters that relate the data to the model
– There are no miracles
Great Bustard
Veteran Member
- Messages
- 45,961
- Solutions
- 17
- Reaction score
- 34,046
It would be the same regardless of the pixel count.
Showing less variation at a lower spatial frequency does not imply greater accuracy.
Except smaller pixels minimize the problem, inasmuch as the problem is recording an accurate image of the scene. The advantage of larger pixels is limited to scenes that are inherently low resolution, and even then, this advantage is only noticeable in very low light.
For all other scenes, using noise filtering on a higher resolution photo will deliver *significantly* better results than beginning with lower resolution in the first place.
Last edited:
Great Bustard
Veteran Member
- Messages
- 45,961
- Solutions
- 17
- Reaction score
- 34,046
The shot noise will be the same for the same proportion of the photo at the same spatial frequency.
Jinx! :-D
W
wchutt
Guest
Jim,
Yes, Schottky was the first scientist to study and characterize shoot noise. For decades this phenomenon was thought to originate from what goes on in the photodiodes, i.e. measurement uncertainty.
Recently experimental results showed shot noise is not related to the photodiodes but is an inherent property of light. (link ). Schottky was wrong about the source of the signal fluctuations. But Schottky did follow Occam's Razor by invoking the simplest explanation given the data and all of his prior knowledge. He was wrong for the right reasons. This is all most scientists can hope for long after their work is published.
After decades of using the terms shot noise and photon noise, there is no hope the rigorous statistical definition of noise (measurement error or prameter estimate uncertainy) will ever be abandoned. Except for certain experiments, it probably doesn't matter.
I hadn't considered that photon noise really isn't noise until this thread. The signal fluctuations are a state-of-nature. The total model for the raw file information content should include the photon noise in the signal parameter terms, not in the noise parameter terms. It is a coincidence photon noise behaves similarly to measurement noise.
--
– There is no substitute for signal-to-noise in the raw data
– Signal-to-noise can not be improved post facto
– Given a model, there are optimal methods to estimate the parameters that relate the data to the model
– There are no miracles
Yes, Schottky was the first scientist to study and characterize shoot noise. For decades this phenomenon was thought to originate from what goes on in the photodiodes, i.e. measurement uncertainty.
Recently experimental results showed shot noise is not related to the photodiodes but is an inherent property of light. (link ). Schottky was wrong about the source of the signal fluctuations. But Schottky did follow Occam's Razor by invoking the simplest explanation given the data and all of his prior knowledge. He was wrong for the right reasons. This is all most scientists can hope for long after their work is published.
After decades of using the terms shot noise and photon noise, there is no hope the rigorous statistical definition of noise (measurement error or prameter estimate uncertainy) will ever be abandoned. Except for certain experiments, it probably doesn't matter.
I hadn't considered that photon noise really isn't noise until this thread. The signal fluctuations are a state-of-nature. The total model for the raw file information content should include the photon noise in the signal parameter terms, not in the noise parameter terms. It is a coincidence photon noise behaves similarly to measurement noise.
--
– There is no substitute for signal-to-noise in the raw data
– Signal-to-noise can not be improved post facto
– Given a model, there are optimal methods to estimate the parameters that relate the data to the model
– There are no miracles
Similar threads
Keyboard shortcuts
- f
- Forum