I don't understand all of the formulas about senor size and pixels. My question can it get to a point where there are too many pixels on a certain sensor? This may be a dumb question but humor and old guy.
rprouty wrote:
... can it get to a point where there are too many pixels on a certain sensor? ...
Rod
It's the age-old issue of non-consensus; some will say yes, others will say no. To fit more pixels on a given size sensor, clearly the hardware photosites which correspond to pixels have to be smaller. The conventional wisdom is that noise becomes more of an issue as photosite size decreases. So everything is compromises, and people have different opinions - people who have real knowledge on the subject (I'm not one of those) and also people who don't.
Mark Twain said "It is difference of opinion that makes horse races." (He says, as if that relates.)
Some people typically make prints around a particular size, so as long as they have enough pixels to provide the level of detail and noise they need, then more pixels probably won't help much.
Others see more pixels as meaning "I can print bigger" and run in to difficulties associated with capturing and enlarging a given image more than they did before - there, more pixels don't necessarily deliver all the expected improvement.
On a physical level, the pixels have to be big enough for the photon to "get in" and making them smaller than the wavelength of the light will make them stop working. Some P&S sensors are about as close to that limit as feasible...
rprouty wrote:
I don't understand all of the formulas about senor size and pixels
To further add to the issue; sensors don't actually have pixels (picture elements) on them at all. Pixels are what you see on the print or on the monitor; they're the end product of the image capture and processing workflow.
What's on the sensor are sensels (sensor elements). It's common (sadly) to use the term "pixel" for both pixels and sensels, but it's not accurate and can actually lead to confusion because pixel size is variable on a given printer, but sensel size is fixed once a sensor is produced.
You have to take into account the advance of technology. So what once would have been far too many is now not enough
Take the 5D . Once it was 12mp now it's 21mp and has better noise handling but I get the impression that unless the next one is 36mp people will cry
The sensel size / noise argument is usually conducted with the assumption of a static sensitivity, but this is not generally the case when comparing different sensor generations. A smaller sensel can give the same "noise" as a larger one if its sensitivity/capacity is correspondingly higher, and this is what has been occurring in recent years: The combination of improved sensel materials with things like gapless microlenses means that sensors have become both more densely packed AND lower noise. Or as grandma would have said, we have managed to have our cake AND eat it
Sensitivity obviously has a limit though ( the point where even a single photon is measurable) and so must capacity (the saturation photon count for the sensel cannot be infinite). I think we are probably starting to close in on these limits now, which means that in future smaller sensels will translate only into better spacial resolution, not decreased noise also. There is then the physical limit that the sensel must be bigger than the wavelength of the light it is detecting, unless they are able to do something really very clever with frequency shifting (non-linear optics?).
There is of course also the option of totally changing the way the image is collected - a complete re-thinking of the physics of the sensor....
BrianO wrote:
To further add to the issue; sensors don't actually have pixels (picture elements) on them at all. Pixels are what you see on the print or on the monitor; they're the end product of the image capture and processing workflow.
What's on the sensor are sensels (sensor elements). It's common (sadly) to use the term "pixel" for both pixels and sensels, but it's not accurate and can actually lead to confusion because pixel size is variable on a given printer, but sensel size is fixed once a sensor is produced.
You know, I've heard this a number of times here, but I've never run across this distinction between sensels and pixels in my professional life (30 years in optical physics/aerospace using a whole bunch of different focal plane arrays). Anyway, I figured like so many things this was probably just outdated industry slang. Old hacks like me in the end user community say pixel while real pros in the manufacturing and development of imaging sensors must use sensel and wince whenever the aforementioned old hacks misuse pixel. Anyway after reading this thread I decided to look up a few of the websites of manufacturers of the various FPAs we use. Lo and behold, they quote "pixel" dimensions as well in their official specs (no references to sensels to be found). So maybe we folks that say "pixel" are still wrong, but it appears we have plenty of good company.
You have too many pixels if it brings your computer to its knees. However, wait a few years and that won't be a problem.
You may have too many pixels if the sensor pitch is less than one half the size of a line pair at your best lens's 50% MTF. For example, the 36 MPx Nikon D800 out-resolves their best lenses, as far as I can tell. I think they do about 80 line pairs per millimetre at 50 % MTF and that is 12.5 micrometres. That sensor has sensel (pixel) dimensions of 4.9 micrometres: less than half the line pair size, and therefore meeting the Nyquist frequency limitations delineated by the Sampling Theorem.
Thus I expect the D800E to be useless compared to the less expensive D800 because the lenses will act as the Anti-Aliasing filter anyway and therefore there will be a minimum of the false image data generated that happens when the Sampling Theorem is violated.
I think the fact that the 24-70 f/2.8 L II has been updated to such sharpness is a strong signal that Canon intends to release a high megapixel camera soon in the range of 30-36 MPx, conceivably more.
Monito wrote:
the lenses will act as the Anti-Aliasing filter anyway and therefore there will be a minimum of the false image data generated that happens when the Sampling Theorem is violated.
I think the fact that the 24-70 f/2.8 L II has been updated to such sharpness is a strong signal that Canon intends to release a high megapixel camera soon in the range of 30-36 MPx, conceivably more.
Monito wrote:
You have too many pixels if it brings your computer to its knees. However, wait a few years and that won't be a problem.
You may have too many pixels if the sensor pitch is less than one half the size of a line pair at your best lens's 50% MTF. For example, the out-resolves their best lenses, as far as I can tell. I think they do about 80 line pairs per millimetre at 50 % MTF and that is 12.5 micrometres. That sensor has sensel (pixel) dimensions of 4.9 micrometres: less than half the line pair size, and therefore meeting the Nyquist frequency limitations delineated by the Sampling Theorem.
Thus I expect the D800E to be useless compared to the less expensive D800 because the lenses will act as the Anti-Aliasing filter anyway and therefore there will be a minimum of the false image data generated that happens when the Sampling Theorem is violated.
I think the fact that the 24-70 f/2.8 L II has been updated to such sharpness is a strong signal that Canon intends to release a high megapixel camera soon in the range of 30-36 MPx, conceivably more. ...Show more →
I agree with the expectation of a high MP model coming - there are other lenses that have been improved also. However, the math should be redone for 10% and 20% MTF before concluding about too many pixels
AJSJones wrote:
I agree with the expectation of a high MP model coming - there are other lenses that have been improved also. However, the math should be redone for 10% and 20% MTF before concluding about too many pixels
Yes, at 50 % MTF there will be some leakage of signal of finer detail at lower contrast levels reaching the sensor (which may or may not have an AA filter). That should be low enough that aliasing artifacts like moire and jaggies and noise-like false pixels will be of low intensity and less of a problem. Thus even if there is no AA filter the lens would act as a weak AA filter for high res sensors in the 30 MPx plus range; similarly weak to the weak AA filters currently used in DSLRs. Even so, I would not recommend getting a DSLR (or other digital) without an AA filter.
Right: quite a few key lenses have been upgraded in recent years: 85 f/1.2 II, 24 TSE II, 16-35 II, 70-200 f/2.8 IS II, and now the 24-70 II.
WebDog wrote:
How do small sensels handle strong light?? Is the dynamic range the same for one large sensel and many small combined?
The capability to handle light intensity (light per sensor area) can be read directly out of the measured lowest ISO the camera can handle. For current sensors this vary from about ISO 25 to about ISO 200. There is no direct correlation between sensel size and lowest ISO. The best P&S cameras with small sensels handle ISO 80, while some u43 sensors with largers sensels only handle ISO 160. To get really low ISO like 32, there are only a few MF digital backs that can handle it.
vachss wrote:
You know, I've heard this a number of times here, but I've never run across this distinction between sensels and pixels in my professional life (30 years in optical physics/aerospace using a whole bunch of different focal plane arrays).
Probably the biggest distinction between the sensor elements and the actual pixels is that the end result pixels have been interpolated from the Bayer pattern that the sensels read.
rprouty wrote:
I don't understand all of the formulas about senor size and pixels. My question can it get to a point where there are too many pixels on a certain sensor? This may be a dumb question but humor and old guy.
Thanks
Rod
The short answer is that current cameras with more pixels than your old 350D and 5D will allow more detail for sure, and there are not too many pixels in new DSLRs. The issue today is the compromise between pixel count, image noise and processing throughput. You make the choice.
I work with images from the 5Dc (8.2 μm pixel size), 5D2 (6.4 μm), 40D (5.7 μm), 7D (4.3 μm), and G10 (2.3 μm). Considering just the pixel level image quality, I can't see much if any difference between the 5Dc and the 5D2 so that decrease in pixel size was technology advance. Making the jump over to the G10, those pixels fall apart easily in PP and really shouldn't be shot at over ISO 200. Even Canon increased pixel size in the G11 and 12 to 2.7 μm size and the G1X is 4.3 μm, same as the 7D. Comparing the 40D with the 7D is more difficult because while there was some engineering gain, I think the 40D images are better at the pixel level than the 7D. Now, I'll be quick to add here that with proper PP the image or print level results may be better from the 7D.
But, I'll go out on a limb and say that with current technology, the minimum pixel size should be between 4.5 and 5 μm for any given sensor size. That corresponds to 12 - 15 MP on an APS C sized sensor, but scales to about 36 MP on FF if done right.
There are two technologies I'd like to see on a new sensor, both coming from FujiFilm, though not used together. One is using two different sensel sizes on the same sensor to increase dynamic range, as was done with the S-5 Pro, and the other is to use a pseudo-random color array rather than a Bayer array, and doing away with the anti-aliasing filter as is being done with the upcoming X-Pro1.
)
