I should let Brainiac explain since he used the terms, but it sounds as if funnelling refers to a process similar to downres with Photoshop, whereas binning would mean some number of pixels (a power of 2) averaged, so treated like one, with no algorithmic "guesses" about how to represent 64 pixels by 25, or 121 by 49.
David Baldwin wrote:
Pixel binning interests me alot. Not just as a way to keep noise down, but also as a way of extending the life of my existing lenses.
I say this because I suspect megapixel yields over the next couple of years are going to go through the roof, and higher resolution sensors may show up the short comings of my existing lenses. If this occurs one option I assume would be for me to pixel bin my 2010 5D Mk3's 40 megapixel sensor down to 20 megapixels, and then all my old lenses would look good again!
Am I being paranoid or is anyone else thinking like this?...Show more →
Yep, you're paranoid. Unless you only pixel peep or print billboard sized murals, a lack of lens resolution will not be an issue for typical web and print display. Although I love to print 12 x 18, mat and frame, I'm pretty much out of walls. So most of my prints are from 5x7 to 8x12 and look just fine. Another 20MP won't change that. How many of us print huge all the time?
I used 'funnel' because I don't know the right term. I mean capturing all four pixels and averaging the result, rather than just capturing one and binning the data from the rest. Funnelling should give less noise, but be slower than binning.
brainiac wrote:
I used 'funnel' because I don't know the right term. I mean capturing all four pixels and averaging the result, rather than just capturing one and binning the data from the rest. Funnelling should give less noise, but be slower than binning.
Ummmm, "binning' *is* an averaging process.
What makes you think it's any different than what you describe as "funneling" ?
I used to do astrophotography with CCD cameras, back in the mid 90's noise and sensitivity was a major concern and the ability of pixel binning was a solution to increase S/N. (astrophotography is all about long exposure and noise increases as a factor of exposure time).
What was established was that whilst pixel binning can, at best, reach the S/N of the equivalent sensor, the electronic design will not allow to practically reach this theoretical limit (noise is not only present at the sensor level).
Also, if I remember correctly, pixel binning will theoritically reach the same S/N only if the signal of the binned pixels are the same (obviously not very usuful) and in all other cases the resulting S/N will be worse than a non-binned pixel.
This being said, pixel binning still does increase S/N but does not allow to be as good as with larger pixels.
Another method that was applied in astrophotography was to combine many shorter exposures into one. Assuming that noise increases with the exposure time, superposing shorter exposures cut the noise (and average it) while increasing the overall signal. Not sure how this could be applied for the typical short exposure we employ.
Lastly for long exposures you can take a "dark frame" which you substract to your picture. This has the effect of actually eliminate the biggest part of the noise, that's what our cameras do when we select noise reduction for long exposure.
Now I am going a bit OT, but from my experience the results of adding shorter exposures had better results than binning (compared at same resolution) with the benefit of not having to trade resolution for S/N.
DavidP wrote:
Ummmm, "binning' *is* an averaging process.
What makes you think it's any different than what you describe as "funneling" ?
Then my understanding of pixel binning was correct. You could take a 16 MP sensor, bin together blocks of four pixels and average them, resulting in a 4 MP image. You lose detail and gradation that you had, and it doesn't seem that you get any better noise result. Better to take the 16 MP image and use a sophisticated algorithm to reduce it in post processing to 4 MP when that is the size you need. The only advantage to binning in the camera would be smaller files, it seems to me.
ChrisGVE wrote:
...
What was established was that whilst pixel binning can, at best, reach the S/N of the equivalent sensor, the electronic design will not allow to practically reach this theoretical limit (noise is not only present at the sensor level).
Also, if I remember correctly, pixel binning will theoritically reach the same S/N only if the signal of the binned pixels are the same (obviously not very usuful) and in all other cases the resulting S/N will be worse than a non-binned pixel.
Photon wrote:
Then my understanding of pixel binning was correct. You could take a 16 MP sensor, bin together blocks of four pixels and average them, resulting in a 4 MP image. You lose detail and gradation that you had, and it doesn't seem that you get any better noise result.
Is that as simplistic as the algorithm is? Averaging isn't necessarily the "best" method. Let's say you have a white pixel next to a black pixel. Averaged together you get just gray. I guess that's a tradeoff you get for reduced resolution and smaller file sizes...
I'm curious what image reduction algorithm Canon uses in their cameras.
So there is still a gain with binning. But the question remains: is the gain worth loosing that many pixels?
Increasing S/N can be done by increasing S or reducing N. Increasing S can be done by assuming that S' from adjacent pixels does not vary too much (not true when there is a transition => blur). But as N is random, therefore local, N cannot be reduced by using adjacent pixels.
Now one cannot use S' from the adjacent pixels to increase S. All we know from the adjacent pixels is in fact their S'+N'...
From my time at university I (vaguely) remember having worked on X-ray imaging (INTEGRAL satellite), where rather than using an optical system (would not affect X-rays) we were using a mask placed at a distance in front of each pixels, X-ray detectors. By using a fourier transform it is then possible to reconstruct the actual X-ray image (with low resolution though).
I wonder if one could perform the same setup on overlay to the sensor to actually measure the expected luminance of each pixel. If technically feasible it would be possible to increase S/N by correcting the noise (eg if the pixel is supposed to be black and is not, the difference is the noise).
This being said, we already have such a mask in front of each sensors: the bayer layer. Ok I am mumbling... I should stop there, not sure whether this all makes sense
ok, maybe I'm stupid but I read through this thread and still don't really understand. Does sRAW at 1/2 the pixels improve digital noise or not? and how much? Any links to a good visual comparison?
Statistically speaking, per-pixel noise is reduced by one bit (one f-stop) when four pixels are averaged (the arithmetic mean is taken). If neighbors are averaged, the image size in pixels is then 25% of the original. Alternatively, you can average across four frames, and retain the image size: this is called "stacking" in astro. Stacking requires image registration and a static subject. Either averaging approach can be applied with no upper limit from the arithmetic, but the sample size grows exponentially: 4, 16, 64, 256, etc.
Dark-frame subtraction (erroneously called noise reduction in photog-land) is a separate concept, and involves a simple arithmetic subtraction. The target is systematic error, or bias (e.g. purple amp glow, hot pixels). Bias is not noise! The dark frame must be created with the same sampling regimen as the main image, otherwise the noise is reintroduced (although the bias will be gone).
Excuse my confusion in previous messages. I was never sure whether sRaw 'binning' was a question of discarding entirely 3 out of every 4 pixels, rather than averaging them. By funnelling I meant averaging, and by binning I meant discarding. If binning means averaging, then sorry for introducing that confusion.
Per pixel noise is irrelevant: it's the per image noise that you see. Changing per pixel noise by binning does practically nothing to the final image's noise effect. Discarding 3 out of every 4 pixels would produce a much noisier result.
There is no noise advantage (from sRAW) and there can't be. Why? This is why:
Think about it from the point of information: where does sRAW data come from?
Full RAW has all the information that comes from the sensor. sRAW has a _reduced_ amount _generated_ from Full RAW data. Anything that any sRAW file can contain, can be regenerated from an original Full RAW file. There is no magical extra information embedded in the sRAW files.
Short version: If you print an image at a given size (e.g. 15x10 cm or 75x50 cm), a image properly generated from a Full RAW file will always look better or at worst equally good as a properly developed sRAW file....Show more →
SRaw is a demosaiced form. It does not need any more demosaicing, but the color needs to be white balanced and transformed from twe camera's color space in whatever.
The data is stored as a three component JPEG image; two components are subsampled. This means, that there is a green components for each pixel, but there is only one red and one blue component for each pair of pixels (in a row).
As every pixel pair represents eight raw pixels (numerically only), i.e. two red, two blue and four green pixels, every color component value in the sRaw fprmat represents two source raw pixels. The values are kept in 15-bit form; I *guess* that the purpose of this form is to store a precise average of the incorporated pixel values (the added precision covers the extra bit resulting from dividing the sum by two).
In other words the sRaw format contains *close to half* of the original values. I have not done anything with it yet, but I do expect a better quality than downresing in PP could deliver....Show more →
Yeah, but what's noise? Is it the error in measurement? Or the crunchy stuff with off color pixels we don't like?
Canon's enginerds got in a bit of trouble by being sincere about accurately measuring wavelength and getting the analog signal to the A/D converter unchanged.
Nikon knew you could just blast away at unpleasant looking noise post capture. So the photos smooth out and lose detail. So what. Look at those dark areas! Smooooth.
So which type of noise are we talking about? Do you want accurate or pleasant looking?
Me, I've just attached a flux capacitor to my D60 and use the fourth dimension to average out the quantum differences. I actually finished writing this reply before I started.
