[CFWhitman]

The same pixel density in APS-C is not 2/3 of FF; it's about 42% of FF. So for your 12 MP FF D3/D700, the equivalent pixel density in APS-C is about 5 MP, which is (oddly enough) about what you get when you put DX lenses on a FF Nikon.

Now maybe you'd like to start a "5MP is enough" thread, and see how well supported that view is.

You're right. I'm sorry. This is the result of me throwing the number out quickly without taking the time to think what I was saying. APS-C is about two thirds the diagonal of a full frame, not the area; and the area is the dimension that we need for the comparison.

Just to be clear, I'm not saying that no one should need more than 5 MP on APS-C (neither did I mean to say that no one should need more than 8 MP when I spewed out the linear relationship without thinking). I'm just explaining what would be necessary to match the ISO performance. Using the correct relationship brings the point home even better (as is should).

[CFWhitman]

Correct me if i'm wrong, things may hae changed in recent years. I seem to recall Medium format digital backs don't have microlenses. because when they are used in conjunction with large format systems where camera movements like shift and tilt are used microlenses cause disturbing colour shifts at extreme settings. but Medium format backs don't usually go much higher than ISO 800 because the idea behind medium format is High quality images and ISO 100 is the optimal setting.

That is quite possible. Thanks for the info; it is quite interesting.

[CFWhitman]

That's not the actual reason. The reason why FF is better in low-light and always will be by about 1-1/3 stop is because it has more area to collect light. Simple physics.

You're actually correct (assuming the sensor technology between the two is of similar light sensitivity), but you're really saying the same thing as I said without realizing it. It is the area to collect light that makes the difference. Pixel density is just describing the difference in area at the lowest level (the pixel), while you are describing it at a higher level (the sensor as whole).

Expressing it the way you did can help someone to understand the principle, but you have to realize that shrinking the size of the sensor as a whole implies the shrinking of each pixel to maintain the same resolution, and it is at the pixel level that the real action takes place. If you think about it, maintaining the same pixel density on a smaller sensor would be basically the same thing as cutting the sensor to a smaller size. If you could really cut a full frame sensor to a smaller size without ruining it, then would you make the part of it that you cut out have worse ISO performance than it did while it was together? Obviously not. It's the fact that you're trying to get some number of pixels on the smaller sensor approaching the number on the full frame sensor, and thus shrinking the individual pixels, that makes the difference.

[jeffkrol]

You're actually correct (assuming the sensor technology between the two is of similar light sensitivity), but you're really saying the same thing as I said without realizing it. It is the area to collect light that makes the difference. Pixel density is just describing the difference in area at the lowest level (the pixel), while you are describing it at a higher level (the sensor as whole).

No this is wrong. it's NOT pixel density it is sensor size alone.. Same amount of rain falls on a square inch regardless of how many "dots" you draw on the concrete (in general).

---

Last edited by jeffkrol; 03-10-2009 at 03:16 PM.

[Jonson PL]

No this is wrong. it's NOT pixel density it is sensor size alone.. Same amount of rain falls on a square inch regardless of how many "dots" you draw on the concrete (in general).

Hi jeffkrol,
I'm very surprised that you would write this. I often read your posts with great interest. But with mistakes like this, it is no surprise that people long for FF, without contemplating the downsides.

Here are from review of A900 at DPr :
"Relatively high levels of noise at anything over ISO 400 (ISO 6400 is of very, very limited use)

And hey, for the most part this is no bad thing; the Alpha 900 is uniquely approachable for a camera in this class, has the best viewfinder on the market and produces, as long as you don't venture into the higher reaches of the ISO range range too often"

24.6 million effective pixels
2.9 MP/cm² pixel density



Nikon D300 review conclusion at DPr:
"High ISO 3200 perfectly usable (if slightly softer due to NR), ISO 6400 usable for small output

Excellent dynamic range from ISO 200 - 800 (good highlight 'reach'), typical at ISO 1600.


On the inside Nikon has worked hard to deliver both better image quality and better performance; you get usable images up to ISO 3200, extended image parameter control, improved dynamic range, automatic CA removal (which immediately improves the performance of all your lenses), six frames per second continuous shooting (eight with the grip / battery combo), a new AF sensor, AF tracking by color and scene recognition. There are also an almost infinite range of customization options available, everything from how many AF areas are used to the size of the center-weighted metering circle to what happens when you hold the FUNC button and turn the command dial."

12.3 million effective pixels
3.3 MP/cm² pixel density



Personally I would stay far away from Eos 5DMark II, and A900. Should I want a FF, I want it for high Iso potential, like D700 and D3.
The downside to low iso FF cams, (and the reasons that marketing cramps a lot of MP in), is that if you crop the 12.7 MP original Eos 5D heavily, then you wind up with a picture that includes few MP, so lower on detail and resolution than a APS-C cam. This is very old news.



If you put 50 MP into a Eos 1Ds Mark IV, the high Iso is gonna be cr@p. If you put 14 MP into Eos 1D Mark IV, with its APS-H sensor; it is gonna be WAY better than the FF cam.

I'm beginning to be less surprised at the amount of people demanding FF.

Earlier, FF cams always meant bigger pixels than crop cams.
One of the reasons that P&S cams have so poor higher Iso.

The more they are, the less they collect light individually.

Exacly

[ManuH]

If you could really cut a full frame sensor to a smaller size without ruining it, then would you make the part of it that you cut out have worse ISO performance than it did while it was together? Obviously not.

Obviously yes, at the image level the cut will be worse because when you'll want to print you'll have to enlarge it more to get a 13x19 than the FF image, thus magnifying the defects. Pixel density and size is only interesting for pixel-peepers. ISO performance has to be evaluated at the picture level, not the pixel level.

Theorically, if you take a FF sensor and push the pixel density, the overall noise (at the image level) should stay the same (Of course in practice, technology, fill-factor, etc have a play). As you increase the density the overall picture stay the same but the individual pixel degrades in quality. It's logic, they have to share the same real estate. The more they are, the less they collect light individually.

What you will see with pixel density increasing is how the quality is degrading when ISO is raised. Right now the pixels are too big to notice a big difference, even at the pixel level, between let's say ISO 100 and 400 on a K20D.

[ManuH]

If you put 50 MP into a Eos 1Ds Mark IV, the high Iso is gonna be cr@p. If you put 14 MP into Eos 1D Mark IV, with its APS-H sensor; it is gonna be WAY better than the FF cam.

High Pixel density giving worse results is a technical problem not a physical one. Theorically when you have more pixels you just split the signal into smaller parts. In practice, the way the pixel collect light, the waste of space between pixels, etc can all give the high pixel density a bad reputation. But this is a technical problem that has been solved for example in the 50D vs the 40D. The Canon 50D has better ISO performance despite the 50% increase in pixel numbers.

[nanok]

Obviously yes, at the image level the cut will be worse because when you'll want to print you'll have to enlarge it more to get a 13x19 than the FF image, thus magnifying the defects. Pixel density and size is only interesting for pixel-peepers. ISO performance has to be evaluated at the picture level, not the pixel level.

just to be sure "we're clear": with digital, the number of megapixels completely defines how big you can print, the sensor physical size does not mean much, unlike film. this is disregarding dynamic range, noise, and so on (the "irrelevant" picture quality characteristics )

Theorically, if you take a FF sensor and push the pixel density, the overall noise (at the image level) should stay the same (Of course in practice, technology, fill-factor, etc have a play). As you increase the density the overall picture stay the same but the individual pixel degrades in quality. It's logic, they have to share the same real estate. The more they are, the less they collect light individually.

again, trying to clarify (and please somebody correct me if i am wrong, i am venturing on thin ice here ): you would be right, only you must take into account that pixels are "quantum devices", (note that by pixels i mean receptors, not full rgb pixels, but that's besides the point for now) and they have clear thresholds: the lower one, where the signal received is drowned by the noise generated trying to read it (very very roughly speaking), and the upper one when the bucket, quite simply, fills (the receptor is saturated and cannot "count" any further, anything more is just "a lot"). following this, as you stated, individual pixels will perform different if they are smaller: they will, for one, need more "effort" to read what they register, this means -- again very roughly -- a more agressive signal amplification, hence relatively more noise (worse snr).

now, if what you meant is that, for the same , intended output print, if you keep the technology the same, in theory the quality should not vary with pixel density (aka: all you get with higher pixel count is the hope of being able to print bigger/crop more, in ideal conditions, and you lose nothing in exchange), that is i think mostly true, you may be even erring on the safe side (look at the k20d output, compared pixel per pixel -- that's is, the wrong kind of comparison -- with the k10d, i think it is safe to say the k20d wins even in such an unfair test, which is one reason why i keep my opinion that the k20d sensor is a marvelous achievment, and, as a side note, i hope to see it again in a pentax, and maybe developed further), this ofcourse not because higher pixel count enabled better performance, but because technology advances, and the sony sensor we are comparing the new cmos with was a damn dinosaur already (though imho the k10d makes superb use of it anyhow)

so bottom line: we must clearly state what we mean when we compare noise: pixel by pixel, or actual intended output (the first is irrelevant for most practical purposes, imho, thoug it seems to be the most often approach),

and, to answer jeffkroll : no, not if the water you drink is in individual buckets, in that case, the size of the bucket will define: how much water you need so you can actually drink something out of each bucket without the effort being more than it's worth, and how much water can a bucket hold at most (both meaning dynamic range), and, easily overlooked: due to statistics, how many of the droplets will actually end up in the buckets will usually aggravate the issues of the smaller (denser population) buckets.

but if talking purely physics, i have to admit your remark is completely valid. manufacturing issues aside, the easiest way to achieve spectacualrly better image quality would be to increase photosensitive area (as long as the photographer doesn't mind bigger lenses, bigger everything, lenses which, at the same relative aperture, will, indeed, put a greater absolute amount of light into the final image, though spread on a larger area)

What you will see with pixel density increasing is how the quality is degrading when ISO is raised. Right now the pixels are too big to notice a big difference, even at the pixel level, between let's say ISO 100 and 400 on a K20D.

correct, snr (signal to noise ratio) again, however one note: try to check the above statement for various lighting conditions, (hint: look at the shadows), you will see that there is a difference, with _any_ camera or sensor, not only k20d. it is unavoidable, really, and first seen in "shadows" because that's where the worse SNR occurs in any shot (that's why many people would rather overexpose on digital, while still avoiding blow out, and "pull down" during raw convertion, rather than the other way around).

overall, i have a hunch we are all saying the same thing, except we are using different words

summing it up, if i dare:

bigger sensor means (in my personal order of preference):

- lower dof for a given field of view (note: camera movements, as in 4x5 systems --scheimpflug effect -- not taken into account)
- higher pixel count without degrading image quality (hence bigger final output, as in, bigger print)
- bigger and heavier _everything_else_ (mainly lenses, for a given field of view)
- in theory, speaking of "digital" sensors, with the same technology (the infamous "everything else being equal, as it seldom is"), better dynamic range (which also trickles down into better (possible) color depth, unless i am missing something) -- but i have yet to see direct and unarguable proof that this actually makes a difference in final output, noticeable by the human eye, or at least exploitable in post processing to some significant advantage

the rest of advantages often quoted are either derivatives of these, or things which are indirect or historical (or both: like, i can easily find an old excellent wideangle for 35mm, not so for aps-c, thus i want "FF" as i am a wide shooter)

overall, to be honest, i am quite content with the k20d, and i'd like to see more of that in the future. i think the aps-c size was a very good one, price/performance wise, for digital, unlike for film ; this might change soon, as manufacturing bigger sensors will become more and more efficient, but i have yet to be amazed (the 3k bucks a900 would get me thinking, if only the minolta mount would not be long dead and burried for me, with the sony logo on the side of the gravestone; i also doubt sony is making, ahem.. is losing less than about what the shelf price of the camera is, for each unit sold. perhaps k-converting an a900 would be the way to go for some people around here )

[jeffkrol]

Hi jeffkrol,
I'm very surprised that you would write this. I often read your posts with great interest. But with mistakes like this, it is no surprise that people long for FF, without contemplating the downsides.

Here are from review of A900 at DPr :

Thanks, and sorry I disappointed you..
anyways I'd be careful of quoting Phil's tests. some have been (arguably) invalidated...
Downsampling and noise power: News Discussion Forum: Digital Photography Review

Just for fun: Gigapixel sensor:
http://ericfossum.com/Papers/Gigapixel%20Digital%20Film%20Sensor%20Proposal.pdf
Some more pixel stuff........
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html#pixelsize
INDEX:
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html
Oh well bottom line per Mr. Martinec:
Bottom line: Among the important measures of image quality are signal-to-noise ratio of the capture process, and resolution. It was shown that for fixed sensor format, the light collection efficiency per unit area is essentially independent of pixel size, over a huge range of pixel sizes from 2 microns to over 8 microns, and is therefore independent of the number of megapixels. Noise performance per unit area was seen to be only weakly dependent on pixel size. The S/N ratio per unit area is much the same over a wide range of pixel sizes. There is an advantage to big pixels in low light (high ISO) applications, where read noise is an important detractor from image quality, and big pixels currently have lower read noise than aggregations of small pixels of equal area. For low ISO applications, the situation is reversed in current implementations -- if anything, smaller pixels perform somewhat better in terms of S/N ratio (while offering more resolution). A further exploration of these issues can be found on the supplemental page. Rather than having strong dependence on the pixel size, the noise performance instead depends quite strongly on sensor size -- bigger sensors yield higher quality images, by capturing more signal (photons).

The other main measure of image quality is the resolution in line pairs/picture height; it is by definition independent of the sensor size, and depends only on the megapixel count. The more megapixels, the more resolution, up to the limits imposed by the system's optics.

See I prefer Emil's "bottom line" to this stuff. Just gives me a headache (actually now that I look at it it's pretty simple.) :
http://forums.dpreview.com/forums/read.asp?forum=1018&message=31267060
No, if the read noise per pixel is constant then the effective read noise increases by the square root of the pixel density. This is why read noise limits density. The same calculation for shot noise shows that if quantum efficiency is constant then shot noise is independent of pixel density:
Dividing up one large pixel into N small ones:
small pixel signal = large pixel signal /N
small pixel shot noise = large pixel shot noise/sqrt(N)
small pixel read noise = large pixel read noise

After adding the small pixels back together:
summed pixels signal = large pixel signal * N/N
summed pixels shot noise = large pixel shot noise * sqrt(N)/sqrt(N)
summed pixels read noise = large pixel read noise * sqrt(N)

[CFWhitman]

No this is wrong. it's NOT pixel density it is sensor size alone.. Same amount of rain falls on a square inch regardless of how many "dots" you draw on the concrete (in general).

Yes, but the same amount doesn't fall on each dot.

Obviously yes, at the image level the cut will be worse because when you'll want to print you'll have to enlarge it more to get a 13x19 than the FF image, thus magnifying the defects. Pixel density and size is only interesting for pixel-peepers. ISO performance has to be evaluated at the picture level, not the pixel level.

Of course you will not get the same quality if you enlarge it more. I never said there was no trade off, but less resolution (edit: actually a narrower angle of view) is not the same thing as more noise.

Edit: Actually, this ignores another factor. If you change the area of the sensor without changing the pixel density, then yes, more light hits it. What is your reward for the extra light that hits is? A wider field of view, not a higher resolution, and not less noise. In order to take advantage of the added area for more resolution, you need to use a longer focal length. When you switch to a longer focal length, you are stretching out less light to cover more area. In order to compensate for this and actually get more light, you need to enlarge the front element of the lens and make the aperture hole physically larger. This is handled automatically by retaining the same f stop number (that is, you need a larger front element and a larger physical aperture for a 75mm f/1.4 lens than you do for a 50mm f/1.4 lens). So you could also make the argument that it is the larger pieces of glass that you get to use to cover the same angle of view at the same f stop with larger area sensors that make the difference.

Of course the more light you have the less you need large pieces of glass, wide physical apertures, and large pixels/sensors to capture as much of what's there as possible.

overall, i have a hunch we are all saying the same thing, except we are using different words

I think so.

Jeff points out some interesting information in his last post. Glancing over this information, my first impression is (but I may be off base here):
Higher pixel density means more noise per unit area.
Higher pixel density means less noise per pixel.
Higher pixel density has minimal effect on signal per unit area.
Higher pixel density means less signal per pixel.

When signal is low (when there is not much light and high ISO would be used) the more noise per unit area and the less signal per pixel are the key factors and higher pixel density means a low signal to noise ratio (bad high ISO performance).

When signal is high enough (when low ISO would be used) and there is plenty of light to go around to all the available pixels, then less noise per pixel becomes the deciding factor and higher pixel density is mostly just an advantage resulting in an increased signal to noise ratio (good low ISO performance).

(Again, I'm not sure that this is really what goes on.)

[Jonson PL]

High Pixel density giving worse results is a technical problem not a physical one. Theorically when you have more pixels you just split the signal into smaller parts. In practice, the way the pixel collect light, the waste of space between pixels, etc can all give the high pixel density a bad reputation. But this is a technical problem that has been solved for example in the 50D vs the 40D. The Canon 50D has better ISO performance despite the 50% increase in pixel numbers.

I completely agree. But this is why I would prefer that they stay at a MP level, and then work on Dynamic Range, etc. And can get better at high Iso potential instead.

As technology advances, each new generation camera might bring one stop better high Iso. But if you then increase MP, then much of the improvement is lost, (since it will just be the same high iso as last time around, only now better resolution).

Eos 50D is newer technology and therefore also brings improvement in high Iso. But had they stayed at the same MP, then the improvement might have been even more noticeable.


I like your point of evaluating at picture level, and not pixel level.

A great high Iso FF cam, would be a lovely addition to my crop cam.

Thanks, and sorry I disappointed you..
anyways I'd be careful of quoting Phil's tests. some have been (arguably) invalidated...
Downsampling and noise power: News Discussion Forum: Digital Photography Review


Some more pixel stuff........
Noise, Dynamic Range and Bit Depth in Digital SLRs -- page 3
INDEX:
Noise, Dynamic Range and Bit Depth in Digital SLRs
Oh well bottom line per Mr. Martinec:
Bottom line: Among the important measures of image quality are signal-to-noise ratio of the capture process, and resolution. It was shown that for fixed sensor format, the light collection efficiency per unit area is essentially independent of pixel size, over a huge range of pixel sizes from 2 microns to over 8 microns, and is therefore independent of the number of megapixels. Noise performance per unit area was seen to be only weakly dependent on pixel size. The S/N ratio per unit area is much the same over a wide range of pixel sizes. There is an advantage to big pixels in low light (high ISO) applications, where read noise is an important detractor from image quality, and big pixels currently have lower read noise than aggregations of small pixels of equal area. For low ISO applications, the situation is reversed in current implementations -- if anything, smaller pixels perform somewhat better in terms of S/N ratio (while offering more resolution). A further exploration of these issues can be found on the supplemental page. Rather than having strong dependence on the pixel size, the noise performance instead depends quite strongly on sensor size -- bigger sensors yield higher quality images, by capturing more signal (photons).

The other main measure of image quality is the resolution in line pairs/picture height; it is by definition independent of the sensor size, and depends only on the megapixel count. The more megapixels, the more resolution, up to the limits imposed by the system's optics.

See I prefer Emil's "bottom line" to this stuff. Just gives me a headache (actually now that I look at it it's pretty simple.) :
Pixel density and read noise per area: Open Talk Forum: Digital Photography Review
No, if the read noise per pixel is constant then the effective read noise increases by the square root of the pixel density. This is why read noise limits density. The same calculation for shot noise shows that if quantum efficiency is constant then shot noise is independent of pixel density:
Dividing up one large pixel into N small ones:
small pixel signal = large pixel signal /N
small pixel shot noise = large pixel shot noise/sqrt(N)
small pixel read noise = large pixel read noise

After adding the small pixels back together:
summed pixels signal = large pixel signal * N/N
summed pixels shot noise = large pixel shot noise * sqrt(N)/sqrt(N)
summed pixels read noise = large pixel read noise * sqrt(N)

No worries

Others have stated the same, how A900 should be used below Iso 1600 :
http://luminous-landscape.com/reviews/cameras/big-three.shtml

And many working Pro stated how the APS-H sensor in the Eos 1D Mark III, went beyond the original Eos 5D. Simply a matter of newer technology, and built to a higher standard. (But it also came at a price jump, though had about the same Pixel density).


Btw, I still have some links that you provided to RH, from Martinec and Sheeby. I’m looking forward to reading them. But they're quite a mouthful, so haven't chewed my way through them yet. Though very interesting

[nanok]

Obviously yes, at the image level the cut will be worse because when you'll want to print you'll have to enlarge it more to get a 13x19 than the FF image, thus magnifying the defects. Pixel density and size is only interesting for pixel-peepers. ISO performance has to be evaluated at the picture level, not the pixel level.

Theorically, if you take a FF sensor and push the pixel density, the overall noise (at the image level) should stay the same (Of course in practice, technology, fill-factor, etc have a play). As you increase the density the overall picture stay the same but the individual pixel degrades in quality. It's logic, they have to share the same real estate. The more they are, the less they collect light individually.

What you will see with pixel density increasing is how the quality is degrading when ISO is raised. Right now the pixels are too big to notice a big difference, even at the pixel level, between let's say ISO 100 and 400 on a K20D.

Yes, but the same amount doesn't fall on each dot.



Of course you will not get the same quality if you enlarge it more. I never said there was no trade off, but less resolution (edit: actually a narrower angle of view) is not the same thing as more noise.

Edit: Actually, this ignores another factor. If you change the area of the sensor without changing the pixel density, then yes, more light hits it. What is your reward for the extra light that hits is? A wider field of view, not a higher resolution, and not less noise. In order to take advantage of the added area for more resolution, you need to use a longer focal length. When you switch to a longer focal length, you are stretching out less light to cover more area. In order to compensate for this and actually get more light, you need to enlarge the front element of the lens and make the aperture hole physically larger. This is handled automatically by retaining the same f stop number (that is, you need a larger front element and a larger physical aperture for a 75mm f/1.4 lens than you do for a 50mm f/1.4 lens). So you could also make the argument that it is the larger pieces of glass that you get to use to cover the same angle of view at the same f stop with larger area sensors that make the difference.

Of course the more light you have the less you need large pieces of glass, wide physical apertures, and large pixels/sensors to capture as much of what's there as possible.

resorting to f stop/fov is tempting, i was about to fall into that trap too, but i thought again about what jeffkrol said (i had a vague impression from reading past threads that he is not exactly a troll ), and after some consideration it suddenly made sense. let's try to look at it this way:

you have a photosensitive area, we do not care about the individual pixels just now, let's assume there's an infinite number of them, and we can agregate them as we wish (or the other way around:divide the area into pixels as we desire). think of aps-c vs digital p&s (huge difference), to make it more obvious. now, assume we don't care about field of view for now, all we want is to obtain some image on the sensor, so we take a 50mm/2.8 with an image circle very capable of covering both, we assume it is perfect (no focus plane curvature, no vignetting,etc). it is clear that for both sensors, this same lens at the same aperture will deliver the same amount of light per area unit (meaning, the same exposure at the same "iso" will give the same "density" of latent image on the sensors). now comes the issue of useful signal: it is obvious that there will be a smaller absolute amount of information delivered on the fingernail sized p&s sensor as opposed to the big aps-c sensor -- proportional to the absolute area --, so to get the same size of final print, we are actually "stretching" the signal more in the case of the smaller sensor (remember, we are not talking megapixels yet), or, to put it another way, less photons will contribute with information for each area unit of final print, for the same size of print obtained from both sensors, because the density of information (light, not pixels!) per area unit on both sensors is _the same_, so in theory we get less useful information per square inch of print from the smaller sensor, because there simply isn't as much as in the case of the bigger one, to begin with.

now, this, in theory, as i noted, is perfectly valid,and there's no way to deny it. however, in real life, that tends to be of little consequence (difference between various formats at low iso is rarely mindblowing), one reason might be that for practical purposes, there is, really, no difference at "native" iso (of course, there is physically really no such thing: there is still an amplifier, even at "native" iso), because the adc will not suck all the information available from the sensor in larger formats anyway, and on the other hand most of it will be lost in pp anyway (as 14bit per channel , for instance, is not displayable anyway with most current media we use in practice), these are all just practical considerations, it is still true that bigger means better potential, and i guess (any factual proof, anyone?) that some people do take advantage of this better potential from ff vs aps-c in extreme conditions (high iso excluded), but for most intents and purposes, while not going into high iso teritory, i think it is safe enough to say it is hard to state the better snr potential of a bigger sensor is actually making a difference, in day to day shooting.

if going into high iso, it already becomes rather more complicated, as we get into issues of noise caused by "reading" the weak signal and so on, as stated before. trying to look at it simply, the signal is stretched even further at high iso (though, you could say, in another dimension this time, not area, but "depth", sort of a third dimension for imaging), so this is when the better potential of the bigger sensor starts to become obvious and actually be used, you can try to imagine it as printing an X MP image: for small prints, x/2 or x mp will give the same result, but when you start cropping, with x/2 you will run out of pixels, while with x you still have useful information to work with, eventhough you threw a lot of it away already, only, as i said, in the case of iso it's not really area but "depth". going even further, i dare say, if the information from the sensor would be gathered exhaustively and recorded as gathered, the concept of setting iso in-camera would become, technically speaking, pointless (as it would be easy to chose after the fact which part of the "depth" to crop, in post processing).

ahem.. does this make any sense?

[jeffkrol]

Bottom line: High exposure zones and/or high ISO, where photon noise and pre-amplification read noise dominate the noise, are rather insensitive to what ISO is chosen once a choice of exposure is selected and care is taken not to clip highlights. Underexposing by a stop, and doubling the raw values in post-processing (that is, applying exposure compensation), yields the same image quality as 'proper' exposure under these conditions. On the other hand, in lower exposure zones at low ISO, where post-amplification read noise becomes important, the read noise goes down by a bit less than a factor of two (in electrons) when the ISO doubles. In this situation, underexposing by a stop and doubling the raw values in post-processing, yields more noise than proper exposure, particularly in shadows.
Noise, Dynamic Range and Bit Depth in Digital SLRs -- page 3
........
all it takes is better circuitry in the camera to significantly improve the post-amplification read noise.

and my favorite heresy.......
so the photon noise referred to the frame size is rather poor, as it is for any digicam; but the photon noise at fixed spatial scale in microns or millimeters, rivals or betters the 1D3, 1Ds3 and D3. If the FZ50 sensor could be scaled up to the size of full frame, it would indeed rival these cameras for photon shot noise performance; and the resolution -- the 2 micron pixels translate into a 216MP (!) full frame camera (of course, for many applications the actual resolution will be limited by diffraction and lens aberrations). It is currently unclear whether this performance could be maintained as the sensor is scaled up by a factor of nearly five in linear dimension -- practicalities of supporting electronics for the pixels, speed in readout and processing, etc, may make it difficult to maintain the FZ50's performance in a scaled-up version.(NOTE) Nevertheless, what should be clear from the preceding analysis is that there is virtually no difference in photon collecting efficiency over a very wide range of pixel sizes, from 2 microns to over 8 microns. ..........................
(NOTE) The real reason pixel wars are not even more out of control. and a reason companies like to keep the public thinking that fewer pixels are better...they'd have to put real processing power in the cameras.. Of course chip circuitry would need to be magnatudes faster......

[*isteve]

Noise, Dynamic Range and Bit Depth in Digital SLRs -- page 3
........
all it takes is better circuitry in the camera to significantly improve the post-amplification read noise.

Exactly, I understand your premise and agree with your conclusion. This solution will also have a major influence on the DR as well as it will allow the designers (or the RAW software) more lattitude to expose more to the left without losing shadow detail.

Its a shame these discussions always end up looking at pixel PITCH rather than pixel sensitivity. The advances in recent sensors are all focused on increasing the sensitive area and well depth of the individual photosites which, if combined with circuitry improvements, will give you a win-win improvement at both ends of the ISO range. The advances in this area are very interesting but seldom discussed.

Looking at the high ISO performance of the D3X is some indication of whats possible in the next APSC sensor, albeit at a print size lower, but I think this will start to make APSC cameras quite usable for low light work. Another generation on and high ISO noise will be low enough for AF and metering to be the limiter.

I think we are already converging on the old distinctions. All sensors will converge on an "optimum" photosite size and architecture, and larger sensors will only be required for larger prints. Because APSC will be very capable, it will remain the predominant format for portable interchangeable lens cameras and will deliver uninterpolated A3 prints which look amazing. You will only need FF or MF if you want to print A2 sizes or higher at gallery quality.

I do wonder though if the downside of the sensor SR mechanism as employed by Pentax and Sony is its inability to use the camera chassis as a heat sink. It certainly seems to be causing issues with the K20D sensor in live view mode.

[Jonson PL]

(NOTE) Nevertheless, what should be clear from the preceding analysis is that there is virtually no difference in photon collecting efficiency over a very wide range of pixel sizes, from 2 microns to over 8 microns. ..........................

Very interesting.

I'll have to take a look at the articles by Sheeby and Martinec that you earlier have provided.

2 microns, is that the same as 2 MP/cm² pixel density ?

Exactly, I understand your premise and agree with your conclusion. This solution will also have a major influence on the DR as well as it will allow the designers (or the RAW software) more lattitude to expose more to the left without losing shadow detail.

Its a shame these discussions always end up looking at pixel PITCH rather than pixel sensitivity.

So the quality of individual pixels are improved ?

Another thing; as I understood it, the Eos 1Ds Mark II, did not have the same quality high Iso as Eos 5D. It was noisier.

Looking at the high ISO performance of the D3X is some indication of whats possible in the next APSC sensor,

How is this, could you elaborate ? How do they relate ?