Today saw the release of a couple of 12 megapixel digital compacts! Theoretically, these should offer more image detail than any DSLR except the 5D and the 1Ds II. For technical reasons related to dynamic range and noise, they won't, even at base ISO (and they won't come close above base ISO). 10 mp compacts don't look any better than 7 mp models (many are worse), and these 12 mp models almost certainly will be no better than the 10s and the 7s. Compacts have hit a wall, using 7 to 12 mp to produce images that, in ideal lighting and with a short scene contrast range, might rival a 6 mp DSLR. Increase either ISO or scene contrast, and they'll quickly end up producing a less satisfying image than even an old 3 mp Canon D30...

Have DSLRs also hit a similar wall, or are they close? The highest resolution DSLR around today, which also wears the image quality crown at low ISO (1Ds II), was introduced in late 2004. All other pro SLRs, except for the speed-optimized 1D mk III, are 2005 introductions. We've only seen two new DSLRs over $1000 since 2005, plus the M8, and all are specialized cameras (1D III, S5 Pro, M8) (ignoring the D2xs and 30D, both of which are very modest refreshes to existing products). From May 2000 to September 2004, the highest image quality around in a 35mm-sized camera moved from the 3 mp EOS D30, which was equivalent to 35 mm film in some conditions (better at small print sizes, didn't enlarge as well) to the nearly 17 mp 1Ds mk II, which solidly exceeds the image quality not only of 35 mm film, but of medium format as well, in almost all conditions. With similar conditions to those placed on the D30 equalling 35mm film, the 1Ds II approaches 4x5...

From September 2004 to May 2007, the image quality bar has not moved at all. Does anyone really think that 18 months from now (late 2008), we'll have a DSLR that has image quality as much superior to the 1Ds II as the 1Ds II is above the D30? It would have to be roughly 85 megapixels, with about 1.5 stops more dynamic range than the 1Ds II, and a nearly noiseless ISO 1600, with a very usable 3200 and perhaps 6400. It would exceed the quality of 4x5 film in all circumstances, challenge 8x10 in most, and be equivalent to 11x14 in certain conditions. Most predictions I'm seeing for the 1Ds III and the Nikon D3 (or D3x) are more like a 22 mp camera, perhaps with an extra half stop of dynamic range. The ISO possibilities will improve, possibly by as much as a stop or two (in this regard, it may be close to as big a step as the 1Ds II was).

There were at least three generations of image quality between the D30 and the 1Ds II - the 6 megapixel group (Canon D60, Nikon D100, etc...), the original 1Ds and the 1Ds II. Over the next equivalent time period, by December 2008, instead of three generations of image quality , we will see only one (1Ds II to 1Ds III/D3), and it will be perhaps less significant than any of the preceding three. The 1Ds III and D3 (or D3x) will both be quite new (less than a year of real availability) at the end of 2008, assuming that the D3x has appeared at all (if Nikon follows the same path (D2h first, D2x over a year later) with the D3 line that they did with the D2 line, the D3x will be touch and go to appear in 2008 at all).

Why has progress slowed so much? I can think of four possible reasons...

1. What do you do with a file that size, anyway? The 1Ds II already makes gorgeous 20x30 prints of complex landscapes. At some point, you don't need more resolution, because you have to step back to see the whole image. The only way to see the extra detail is either by pixel-peeping on screen or by examining a very large print unnaturally closely. I don't know if the 1Ds II is over that line, but it's very close to it.

2. Have the laws of physics reared their ugly head? Compact camera makers are willing to introduce cameras with more megapixels, but equivalent or even inferior image quality, because the megapixels look good on marketing posters. That doesn't work for pro DSLRs! Most people who buy a $5000 - $8000 camera want to see that it really does look better on a print than what they have!!! Most buyers of these cameras are very experienced photographers (professionals, artists and longtime amateurs) who know what they like and don't about their current camera's images, and buy something that does better at what they don't like.
In my own case, I switched systems from Nikon (D200) to Canon (1Ds II) looking for resolution, but primarily for shadow detail. I got roughly twice the resolution (the 1Ds II is a little better per pixel, plus it has 1.7 times as many pixels), and I'd say an extra stop in the shadows, which was what I was looking for for my landscape work. Because most of us know what we're looking for, neither Canon nor Nikon will release a new camera in this class that doesn't offer REAL improvements in image quality over anything they already offer. It is possible that the 22+ mp sensors that fit with 35mm lensmounts, which both makers are experimenting with, are enough noisier than today's sensors that the images are actually no better. If that is the case, it could either be a simple engineering issue (a better sensor is possible, but we're not quite there yet) or an actual physical barrier (decreasing pixel size will inevitably add noise - we've hit the limit). If we're close to, but not quite at a physical barrier, each step closer to the barrier will be harder and take more time than the one before...

3. Is it the lenses? The 1Ds II already has a reputation for out-resolving lenses, especially wide-angles, and especially at the corners. Is it possible that any higher resolution camera simply out-resolves most existing lenses, and offers no meaningful improvement in image quality because of it?

4. Is it the marketing department? A new high-end SLR is a huge commitment of research and development funds, especially if it will use an entirely new image sensor. A 1Ds mk IIn might not be, if it was simply the existing mk II sensor (or a slight improvement at the same resolution with lessons learned from the 5D) in the 1D mk III body, and I'm sure Canon will introduce that soon if they don't have anything else in the pipeline (it would give the 1Ds line 5 fps, improved buffers, better AF and a better LCD screen). Anything more than that, from either Canon or Nikon (even more so from Sony or anyone else, as only Canon and Nikon have the lenses to support this sort of camera), is an investment of millions of dollars for a camera that sells a few thousand units per month. The prices are high enough that the camera will eventually turn a profit, but is it AS profitable as building yet another vehicle for the ubiquitous 10 mp Sony sensor (anyone except Canon), or for a 20D or XTi sensor (Canon) - much lower R and D, many more sales (at a lower price)...

What do other folks think?

-Dan

Really interesting.

I would say it is the following:

Canon is still alone with the 1DsMk2, there is nothing which comes near ir or is FF.
So they have all time in the world.

Nikon is doing nothing or can't do nothing because they can't make their own sensor.
So, they have to wait for Sony, and Sony was never a fast company.

--> Canon is alone and can wait until any other company moves.

This is so sad but that's how it works, as soon as there would be 2-3 companies pushing the limits, we really would see improvment.

This is the same for all markets, from CPUs to cars, as soon as one company is allone it's the death of progress. Very best example M$ and Windows.

Dan,
I wouldn't think so. The 400D pushed the envelope by narrowing the gap between microlenses. This allowed noise and DR to remain pretty close to that of the more expensive 30D.

Whislt it's true that the differences in resolution between the 10mp 400D and the 8mp 30D might be so marginal as to be almost irrelevant; and whilst it's also true that the differences in resolution between the 8mp 20D and 6mp 10D are similarly insignificant (but probably not quite as insignificant), the resolution differences between the 400D and 10D are almost certainly worthwhile.

The 1Ds MkII has the same pixel density as the older 10D, which means that the 10D has no resolution advantage whatsover in any circumstances with any lens, whatever the degree of cropping. The 1Ds2, however, has an advantage of lower noise in most circumstances. Technology marches on.

A full frame sensor as in the 1Ds2, but having the pixel density of the 400D, would be around 26mp.

By the time Canon is ready to introduce such a camera, it is reasonable to suppose there will have been further technological improvements that will ensure that, on a pixel for pixel basis, those 26mp will have at least the same performance of current 1Ds2 pixels.

Anyone who owns both a 1Ds2 and 400D should be able to get an idea of what to expect when they compare 1Ds2 shots cropped to the same format as 400D shots using the same lens. If the 400D shots looks any worse with regard to non-resolution factors, such as noise, dynamic range, tonality etc., these will be addresed and fixed before such a 25mp camera is released to market.

However, this scenario I've described will probably not eventuate for a number of years. We now have a 10mp 1D MkII with improved noise characteristics, despite a marginal increase in pixel density.

I find it difficult to believe that Canon's most expensive, flagship model would ever have worse noise than the 1D2 and a lower maximum ISO capability, so we're probably looking at 20-22mp for the 1Ds3, and 26-30mp for the following upgrade.

On the other hand, we can't assume that technological progress will just continue with the same paradigm. Once one paradigm is exhausted, a new pradigm often takes over which offers further improvement and sometimes a dramatic jump in improvement.

Ray suggests with fairly good evidence that a full-frame sensor in the mid-20 megapixel range is technically possible without losing image quality to undersized pixels.

Will we be able to see its advantages in prints?

Will we be able to make lenses that take advantage of it?

Will the marketing departments permit it to be made - is there enough profit in it to interest a big camera maker?

-Dan

First an aside: the new 12MP sensors seem not to have smaller photosites than previous 10MP ones: the sensors are slightly larger, 1/1.7" rather than 1/1.8". But still, the main point is that digicam sensors seem fairly close to giving the maximum resolution possible due to lens limitations like diffraction, and are probably already at or beyond the resolution needs of almost every print that will ever be made from them.

I am inclined to think that for all but very high end photography with very good lenses, DLSR's are also not so far from their resolution limits, partly due to the declining demand for each further significant increment in pixel count, due to the IQ differences being visible only at ever larger, less common print sizes. Going further is technically possible, but probably only at the cost of a very expensive lens collection, which only the very high end will bother with. And the low sales volume of such high end gear will make those lenses even more expensive, in the way that medium format lenses are usually far more expensive than comparable 35mm format lenses. (Compare prices of telephoto lenses 200mm and up for 35mm and MF cameras: at these focal length, the larger format does not require a more difficult design, as the natural image circle size of a 200mm or longer lens design is already easily big enough for MF.)

So, though I can imagine the high end of 35mm format DSLR's going as high as 30MP, and Kodak/Dalsa have reportedly talked of going to about 60MP in 36x48mm, I expect the vast majority of DSLR's and even the majority of DSLR gear used professionally to stay below 20MP. Maybe well below.

About professional DSLR needs: I believe that Canon's 1D models have always been its best selling "pro" models, produced in about twice the numbers of the 1Ds models, and Canon has decided to take the 1D MkIII to only a modestly higher 10MP (and to keep it at 18x27mm format). An ironical contrast to 12MP digicams.

BJL,
You have a point there, but there are still a number of high quality and affordable primes and zooms which seriously out-resolve the current FF DSLRs, at f8 at least.

The humble Canon 50/1.8 II costs hardly more than a basket of groceries, yet it's razor sharp at f8, if you are lucky enough to have a good copy, as I have.

I think the Canon 70-200/2.8 and 70-200/4 zooms, at f8, are also sufficient to provide greater resolution than current FF DSLRs can offer, and are still reasonably affordable, especially the 70-200/f4 IS.

As a matter of interest, when recently perusing the Photodo lens archive in search of lenses that came close to the theorectical diffraction limited resolution at f8, the Canon 70-200/f2.8 non-IS was the only lens I noticed that had 80% MTF at 40 lp/mm. However, this spectacular result was achieved only dead centre at a specific FL which I can't recall offhand.

All lenses should ship with their own, comprehensive MTF charts. Those who don't understand them should be impressed by the fact they have been provided. Those who do understand them will find them extraordinarilly useful, not only to choose the precise FL and f stop for the sharpest result without a lot of tedious testing, but also as an aid in choosing the right lens to buy.

The downside would be, we would not have the opportunity to have endless debates about this lens being better than that. Instead, the focus would be on discrepancies between the accompanying MTF charts for a particular lens and the actual performance. So maybe it wouldn't be as much fun because there would always be the possibility of a resolution.

I agree that a few affordable lenses are available with enough resolution to make substantially higher pixel counts than 12MP or 16MP worthwhile. But that is what I used the words "lens system". The photographers willing to limit themselves to primes are probably a quite small fraction of SLR users, and roughly the same as those who accept the limitations of medium format film systems for advantages in resolution and such. For the rest, the extra resolution being worthwhile with good zoom lenses is probably an important criterion in judging the worth of a pixel size decrease.

So perhaps a key question is how much sensor resolution is worth having with lenses like the main line of Canon L zooms: 16-35/2.8, 24-70/2.8, 70-200/2.8 or 17-40/4, 24-105/4, 70-200/4. At your favorite aperture of f/8, MTF of 50% or better is limited by diffraction alone to 95lp/mm, so maybe there is a natural limit on worthwhile sensor resolution of 100lp/mm or less.

However, that ignores the economic question of how many photographers will find significant extra value in going much beyond say 16MP, given that only rather large prints or unusually close scrutiny will reveal a difference. Maybe the resolution, pricing and sales of the new 35mm format DSLRs expected later this year will give us a better idea.

The 1DsII already has a resolution of nearly 70 line pairs/mm, and the D2x is close to 90! This explains why lenses for these things are such a problem... The D2x is diffraction limited just past f8! A full frame camera with the pixel pitch of the D2x is just under 28 mp, and that is surely the limit (if nothing else, who wants to deal with a camera that is diffraction limited at f5.6)?
The medium format backs are actually pushing the limits at least as closely right now (I've never heard of a medium format lens that goes beyond 80 lp/mm, and 80 lp/mm at 36x48 mm is 44.2 mp - the P45 is awfully close).
If no other gremlin is going to reach out and grab us, diffraction is a good candidate! Right now, full frame and larger cameras are diffraction limited at f11, and crop-sensor cameras are limited at f8. Diffraction is basic physics - no amount of fancy technology can save us from its effects. Due to the inherently greater depth of field of a crop-sensor camera, a cropped camera limited at f8 is somewhat more flexible than a full-frame model also limited at f8. A few primes and one or two long zooms (the 70-200 f2.8 and f4, in both Canon and Nikon versions) have the resolution to hit the diffraction limit at f8, and maybe even f5.6, but is a camera that delivers its maximum resolution with a few lenses, at with a one-stop aperture range (f5.6 to f8) all that useful? If a single-aperture camera (f5.6 only) is acceptable, the diffraction limit limits a full-frame camera to 70 mp, and a DX model to 32. This means that the camera that is as big a step above the 1DsII as the 1DsII was from the D30 is impossible to build...
As an interesting aside, the new 12 mp point and shoots are diffraction limited WIDE OPEN as soon as you move the lens off of maximum wide-angle. The diffraction limit is f3, and the lenses are f2.8 at maximum wide-angle, going to f5.6 at full telephoto. The next step is for some manufacturer to build a camera with an inherently useless resolution - there is NO aperture anywhere in the zoom range where the camera isn't over the diffraction limit. There may be a few cheapo models where this is true, but I have yet to see one from a major manufacturer. There is a good chance that most of the lenses on these compacts aren't up to the task, though. They're VERY high resolution lenses (well over 100 lp/mm), an optical design feat made easier by the fact that the sensors are tiny - it's easier to get huge resolution over a small area. However, the highest resolution sensors in these compacts are theoretically capable of over 250 lp/mm. What are the chances that a $250 compact actually has a 250 lp/mm lens? Does a 250 lp/mm lens even exist?

-Dan

There will be a way and companies will find it. So I'm sure we will see more MP and more and more, hopefully soon.

Dan,
These resolution figures seem to be theoretical and based on the Nyquist limit of a pixel per line. Dpreview resolution tests show absolute resolution of 50 lp/mm for the 1Ds2 and 64 lp/mm for the D2X.

Even if you disagree with the actual numbers, the resolution differences between the two sensors are significant enough to justify the eventual production of a 28mp FF sensor and even higher. However, with such high pixel count FF sensors there's always going to be a problem in the corners of the frame in situations where good DoF is required, which includes many landscape shots. This seems to me to be the downside of high pixel count FF sensors. The resolution of the recorded image will more closely resemble the actual resolution of the lens. The vast majority of 35mm lenses exhibit serious resolution fall-off starting about 18mm from the centre, and many even before that.

Basically, we could say that full frame sensors have already reached and even gone beyond the resolving power of most lenses in the corner of the frame.

On the other hand, the corners of the frame are not high priority for most shots.

Yes, the numbers I posted are theoretical, using a perfect lens - of course the camera won't ever reach them with a lens on it. However, the dpreview test you mention has its own flaws - they don't mention what lens was used (dpreview is reputable enough that I'd guess it's a decent lens, but I'd like to know). They also used a sharpening setting of 0 on the 1DsII resolution chart - the 1DsII is known to require some sharpening, and to handle it well. Not knowing the D2x, I don't know if it is possible to turn the sharpening entirely off in a JPEG (it IS possible on the 1DsII, and dpreview did it). Fixing the sharpening issue may increase the 1DsII's overall resolution lead, and perhaps tighten the practical lp/mm. On the other hand, the Nikon could have been tested with a less capable lens than the Canon (therefore narrowing the overall gap and widening the Nikon's per mm lead). I don't know how to do a good practical resolution test across lens mounts (within one mount, using the same lens and processing a RAW optimally in the same software seems the fairest method). Across both mounts and sensor sizes, there's probably no fair way to compare, but lenses of similar high quality and close field of view (50 f1.4 Nikkor against 85 1.4 Canon?), and a RAW converter known to do a good job on both cameras' files (Capture One?) is probably as fair as anything...

-Dan

DPReview usually use a 50mm lens. For their Canon tests, they use the EF 50mm f/1.4.

And as far as I know, they use the same lens for the resolution chart tests as they use for their noise tests.

Although the EXIF data for the images do not include the lens name, they do show the focal length used; 50mm.

I would not get too hung up on the diffraction limit at f/8 when considering the resolution limits of lenses for use with smaller than 35mm DSLR formats like DX or 4/3. The lens tests at sites like http://www.photozone.de/8Reviews/ show that Nikon DX and Canon EF-S lenses often have peak resolution at around f/4, higher than what they show at f/8. Similarly for Olympus 4/3 lenses, though the 11-22 wide angle zoom at 11mm and the 50/2 ar as sharp at f/2.8 as at any higher f-stop.

A number of those test reports also say that the measured resolution is limited primarily by the sensor, not the lens.

If diffraction at f/4 instead of f/8 is the rough limit, pixel sizes down to significantly below 4 microns could give worthwhile gains in resolution, so DX is probably not close to its lens-based resolution limits with its current 6 micron (10MP) and 5.5 micron (12.4MP) pixel spacing; likewise EF-S and 4/3 seem to be not yet that close to the resolution limits of their lenses.


Note that halving the diffraction limited f-stop doubles the diffraction limited resolution and roughly quadruples the diffraction based upper limit on useful pixel count.

P. S. The Canon 70-200 f/4 and f/2.8 L lenses that Ray suggests as excellent examples of good resolution at large apertures mostly have peak resolution at f/5.6, but ranging from f/4 to f/8 for some choices of lens and focal length, and the same is true for the Nikon 70-200/2.8.

BJL,
I can't find a definition of centre resolution in those Photozone tests you refer to. Are we talking about 'dead centre' or a centre with a diameter of say 9mm or 12mm etc?

Whenever I've tested my lenses with resolution charts (and various textures I've stuck of the board), I've found it impossible (so far) to find any part of the board that is actually sharper (or even as sharp) at apertures wider than f8. Maybe I'll have to redesign my test board and stick a 10 cent mint condition coin in the centre.

The general impression I get is that over-all performance of any 35mm lens is rarely better than it is at f8. On the other hand, without the accuracy of 1 series Canon camera, perhaps my focussing is always slightly off.

I prefer the old-fashioned Photodo MTF charts to these 50% MTF Imatest charts because (with the MTF charts) you get effectively a running commentary on the varying performance from centre to corner of the lens itself. Unfortunately, Photodo did not test their lenses at every f stop so it's not clear at what f stop a particular lens might have peaked in the centre.

Consider the following Photodo MTF charts for the Canon 70-200/f4 USM (non-IS).

Click to view attachment

What's remarkable here is that, at 200mm and f4, this zoom achieves the theoretical diffraction limited resolution of a lens at f8, ie. 80% MTF at 40 lp/mm. In the centre, resolution is actually higher at f4 than it is at f8, but over-all resolution at f4 gets a 'just marginally' lower rating of 0.81 than the f8 rating of 0.82. If it were not for the fact that photodo made allowances for poor performance towards the edges (ie. it's a weighted result which attaches less importance to corner performance), the rating at f4 would be lower.

Nevertheless, such small differences are probably irrelevant. The point here is, just because a lens peaks in the centre at f5.6 or f4 is no indication the lens is diffraction limited at these f stops. In fact, nothing could be further from the truth.

I do not see how that is particularly relevant to the basic significance of the Imatest results. By the way, off-center figures oft have the same result, peaking at f/5.6 or f/4.


This sounds an awful lot like trying to ignore a body of seemingly careful experimental evidence simply because it does not support your prior beliefs. Not good behavior for a self-described "rational fundamentalist"!



Actually, the point here is not at which aperture a lens is or is not diffraction limited: the point is estimating the maximum resolution possible from a lens, regardless of what mixture of diffraction and aberrations contribute to that lens resolution level. This maximum lens resolution is what matters to questions about how high it is worth pushing sensor resolution for a given format and lens system.

f-stop for peak 50%MTF resolution is not the ideal measure, but I know of no better data for now. Certainly MTF at 40lp/mm and at only f/4 and f/8 are not much of a measure of maximum resolution when the latter is clearly at far more than 40lp/mm, and likely occurs at neither f/4 nor f/8 but somewhere in between. If only photodo had f/5.6 data, we might see MTF better than the f/8 diffraction limit. At least for some DX, EF_S and 4/3 lenses!

By the way, am I reading right that f/4 does better than f/8 at 70mm and 200mm for peak (on-axis) MTF in those Photodo graphs? Since I can see a value to sensors getting the most out of a lens' optimal center of field sharpness even if the sensor then "out-resolves" the lens towards the edges of the image I care mostly about that peak, on-axis resolution.

The sensor of the 1d3 seems to be at least one stop better in terms of noise than its 3 years old predecessor everything elese being equal. It does also probably have a better DR. These 2 aspects seem to show that digital photography is no different than any other consumer electronic field.

It seems pretty obvious to me that technology is more of a limiting factor than physics at this point of time, and we, non-experts, have no way of figuring out how close we are to the "theoretical wall" that has been predicted again and again. By the way, there are NO real experts out there besides those actually working on sensor design...

We see very often in most fields of consumer electronic companies coming up with products jumping a few generations ahead of their competition.

If Canon could outdo itself between the 1d3 and the 1d2, I see no reason whatsoever to think that Sony, Samsung, Kodak or any other of these electronic giants cannot produce something significantly better in terms of sensor, meaning something nearly as good with a much smaller sensor...

My personnal assessment is that my 200 gr. Ricoh GX100 already produces most of the time better images than my old F100 with a Nikkor 28-70 f2.8 and Provia 100F loaded, while having a much larger envelope of typical usage, both in terms of lens range and speed/aperture combos... Facts show that the smaller the sensor, the better the lens become. Diffraction is less of an issue with smaller sensor than with larger ones because DoF appears to grow faster than diffraction as size is reduced.

I feel that it will only take another 3-4 years until compact digital cameras provide an image quality sufficient for 99% of amateurs's usage, and at least 50% of pro usage.

The reason why people will keep using DSLRs even then has more to do with interface, robustness, ability to have shallow DoF,... in other words all these reasons that led people to work with DSLR in the film days already.

Sensors are nearly already a commodity.

Regards,
Bernard

BJL,
Good try! I see you read my unfortunate post in the 'About this Site'.

I don't think it's true I've started off with prior beliefs on this issue, ie. at which aperture a 35mm lens is sharpest, although I've always been aware that stopping down below f11 can reduce resolution.

In fact, I never even bothered doing resolution tests before switching to digital photography which has the advantage of allowing for very quick feed-back on results. As I mentioned, I haven't paid particular attention to resolution in the dead centre, so next time I do some testing I'm going to place something appropriate exactly in the centre of my test board.

If that central area where a lens is sharpest is very small, then it's of limited usefulness. For example, when photographing people or wildlife, one often tries to get an eye in perfect focus. If the lens used is sharpest at f4, but only in the centre, then probably more often than not results might be disappointing, except in those circumstances where the composition dictates that the eye is also in the centre of the frame.



Whilst it's true that Photodo MTF results would have been an even better guide had they tested at more f stops, they usually provide weighted ratings for the intermediate stops between f8 and full aperture and such ratings are, in my view, a better guide than two vertical bars which purport to describe average performance over an unspecified portion of an image circle. But maybe that area is specified somewhere. It just seems to me to be a very significant piece of data that is either missing, glossed over or difficult to find.

For some reason, Photodo tests do not seem to provide weighted ratings at f5.6, except where f5.6 is the maximum aperture. I would guess that the reason for this is that weighted results at f5.6 have been found to be in most cases virtually the same as at f8.



This is a situation which highlights the advantages of the cropped 35mm format. Only in the corners of the APS-C format would f8 show better resolution and I would say in general that at f4, where DoF will likely be shallow, corner resolution is likely to be a non-issue for both cropped format and full frame.

It's no wonder this lens has a fine reputation . The IS version will probably be my next lens purchase.

Even if we have a significant number of lenses which have extremely high corner-to-corner sharpness at f4, and are capable of using a sensor's full resolution at an aperture that wide, we still have a diffraction issue. Yes, the diffraction limit at f4 permits a huge number of pixels (around 61 mp on a DX sized sensor, and 140 mp on a full-frame sensor), but that means that the lens can ONLY be used at f4 and wider (in the case of the exceptionally rare lens that can actually deliver that kind of performance at f2.8) for maximum performance, severely limiting depth of field. To have a significant number of aperture choices available, permitting creative control over depth of field, I would argue (and this is purely subjective) that f8 absolutely needs to be accessible on a full-frame sensor, and f11 is important, if not truly essential. Reduce those numbers by one stop (to f5.6 and f8) for a DX camera, due to its inherently greater depth of field. To have a non diffraction limited f5.6 on DX, and f8 on full-frame, maximum resolution is in the low 30s of megapixels (in both cases - the depth of field advantage cancels out the diffraction disadvantage). To preserve one more stop of creative control, retaining f8 on DX and f11 on full-frame, we are limited to 15 mp on DX and 18.6 mp on full-frame.

-Dan

Dan, this seems an ambiguous phrase (in bold). A sensor's full resolution might be considered as a pixel per line width. Apparently, only the Foveon type sensors achieve this. Most DSLRs have Bayer type sensors which have to interpolate color values, so it's not really the fault of the lens that most sensors do not achieve full resolution up to the Nyquist limit. There's also usually an AA filter in front of most Bayer type sensors which softens the image perhaps in a similar way to stopping down.

I recently compared my Canon 50/1.8 II, 50/1.4 and 24-105 zoom at 50mm, at various apertures, shooting various test charts and textures I'd stuck to a board. I used my highest resolving camera, the 20D for this purpose even though my 5D is now the preferred camera for most occasions.

The purpose was to find out if there would be a more noticeable difference between resolution at f8 and f16 if I were to use a sharp lens, and just how noticeable and significant such differences really are. While I was at it, I also shot at all apertures up to the maximum aperture of each lens.

Unfortunately, I didn't have the appropriate section of a test chart in the dead centre of the frame, with varying line widths on either side of the resolution limits, so I cannot examine critical results closer than 4mm from the centre.

Having examined the results again at 200% on the monitor (which represents a huge print size), I cannot find any instances where resolution at f5.6 is noticeably better at a distance of 4mm-5mm from the centre of the frame. The best I can say is that on average, resolution at f5.6 and f8 are about equal.

Below is a comparison using my sharpest lens, the Canon 50/1.8 II. To show just how close resolution is at f5.6 and f8, I've captured on screen views at 400% and compressed using maximum jpeg quality. If you want to put a really, really fine point on it, the f8 shot is very marginally better in my opinion. I went to some trouble to define the exact centre of the frame. The point where resolution breaks up is about 4mm from the centre, give or take 1/10th of an mm.

Click to view attachment



It's always going to be the case that, as lens aberrations are reduced and pixel count is increased (on the same size sensor) perceived DoF on a sufficiently large print will inevitably become shallower at f stops that were previously not diffraction limited. You would expect a lens that is diffraction limited at f4 to have twice the resolving power as it does at f8. Excluding such factors as AA filters, a sensor of any given size that cannot capture resolution greater than a diffraction limited f8, would need to have 4x the number of pixels to capture such an increase in resolution. DoF on a sufficiently large print, at f4, might then appear about the same as it would with current lenses at f1.4.

BJL might like to comment on this relationship.

P.S. I should mention that anyone looking at the above resolution test should not draw the conclusion that the 20D can resolve over 70 lp/mm, as the charts seem to imply. In oder to determine the precise resolution limits, one has to position a chart of a particular size at a precise distance from a lens of a particular focal length. I haven't done this in these circumstances. I'm merely making a comparison rather than trying to determine absolute values.

One comparison I care about is increasing sensor resolution while keeping everything else the same. A lens can always be used at a smaller, "diffraction limited" aperture, regardless of how high sensor resolution gets, with the overall resolution being at least as high (probably a bit higher) than if the same lens and aperture are used with a sensor of lower resolution. The various factor limiting resolution are not brick walls, but combine to give overall resolution lower than the limit due to any one factor alone.

The worst that can happen to image sharpness and detail is that higher resolution only gives a useful improvement in resolution in a very limited range of cases, of rather large apertures and low DOF (with low DOF made more noticeable by the greater enlargement and or closer viewing used to see the extra detail.) Not that performance is worse than with a lower resolution sensor at small apertures, or that small apertures cannot be used at all.

My skeptical expectation about increasing sensor resolution "too far for the lenses" is just ever smaller improvements in image quality, possible only in ever more limited circumstances (such as only with ever lower DOF) so that at some point, the "law of diminishing returns" will make it economically non-viable to push sensor resolution further. Ever fewer customers will see the ever more limited benefits as sufficient to justify the ever greater costs (better lenses an such) and ever greater effort involved in achieving those benefits.

I agree that using a sensor at apertures above its diffraction limit will not provide any WORSE results than using a lower resolution sensor running at (or just below) the limit. In addition to the development cost that BJL mentions, there are is another limit to this approach, though. The second issue is that higher-resolution sensors are often not equal to lower resolution ones in other aspects of image quality (such as noise and dynamic range). This will become more acute if sensors with extremely high performance other than resolution (notably Fuji's SuperCCD, potentially also the Foveon three-layer CMOS or some relative) show up in more competiitive resolutions. Both of these unusual designs have some performance advantages, but both are hampered by notably low resolution at present (the Foveon is also hampered by low-end Sigma bodies with a nonstandard lens mount, while the SuperCCD was hampered by body problems until the S5). Imagine the choice between a D3x with a 22 mp conventional Nikon DX sensor, diffraction limited around f6.3, and a Fuji S6 with a 1.2x crop SuperCCD with 14+14 mp. The Nikon sensor clearly outresolves the Fuji, but at a very limited aperture range, and the Fuji has MUCH better dynamic range. I'd take that "S6" over the "D3x" in a heartbeat, given those specifications. The S6 even offers some advantages over a 1Ds mk III with a 22mp full-frame CMOS sensor. The delay in the 1Ds mkIII may be caused by Canon trying to find a way to make a 22mp sensor with per-pixel performance (noise and/or DR)equal to their existing 17mp sensor.

-Dan

Dan,
It would be fascinating to find out what ideas and possibilities the guys in Canon's labs are currently working on. Is there an unusual delay between the 1Ds2 and 1Ds3? We have only one precedent here and that's the interval between the 1Ds and 1Ds2. How long was that? I don't see that Canon is under any obligation to bring out new models in a clockwork manner. Also, it doesn't seem good marketing policy to have too many goodies competing with each other for the consumers' dollar. It tends to dilute the return on R&D and setup costs.



This only applies when all else remains the same. Perhaps it's odd that no manufacturer seems to have pulled out all the stops to produce a high DR camera at the expense of resolution. Would there be a market for a 6mp full frame 35mm sensor with 16 bit processing and an extra couple of f/stops of DR?

The current level of dynamic range in DSLRs is roughly that of color negative film. Most photographers seem reasonably happy with it, but I agree that Canon does not want to go backwards in this area.

I can't see why there should be an insurmountable obstacle to producing say a 50mp 35mm sensor with noise and dynamic range characteristics at least as good as the current 1Ds2. For dynamic range and noise to remain the same on the same size print, it is not necessary for the smaller pixel to have the same low noise and the same DR as the larger pixel. All that is necessary is to keep noise and DR the same per unit area of sensor, and this could be achieved by reducing read noise and increasing quantum efficiency.

Let's consider what might be involved with our imaginary 50mp camera which would have a pixel pitch of half that of the current 1Ds2. If the light gathering capacity of the 4 smaller pixels in total were equal to that of the one larger pixel and, if the read noise of each of those smaller pixels was no greater than 1/4 the read noise of the larger pixel, then DR and noise would be the same for any given size print, but the 50mp camera would produce higher resolution and more detail on a sufficiently large print.

It might also be possible, as manufacturing processes develop, to place all the photon receptors on one side of the chip and some, if not all, of the processors, such as analog preamplifiers, on the reverse side of the chip. That would allow the photodiodes to be larger within the same pixel pitch.

Has anyone seen a double-sided chip in any application? Interesting idea, and I've seen double-sided (and even multi-layer) circuit boards all over the place, but never heard of an actual chip being double sided...

-Dan

In a sense (although it's not FF), the Fuji S5 is pretty much Ray's "dynamic range at the expense of resolution" camera. It's an $2000 6 MP camera when all other 6 MP SLRs are $500 economy specials. Apparently the dynamic range beats anythine else on the market (including MF backs?)


-Dan

Yes but do you really think they are selling many 1DsMk2 now ? I really don't think so.

Diffraction might be the key point in limiting useful sensor resolution.

Some links with useful reading:

http://luminous-landscape.com/essays/Equiv...s.shtml#another

http://www.cambridgeincolour.com/tutorials...hotography.htm#

http://www.luminous-landscape.com/tutorials/dq.shtml

So far as I have understood it seems valid that from the technical point (year2007) it is not very useful to go beyond 20 MP with a bayer sensor as diffraction increases with the MP count. So the more MP the less useful f stops you have. The only possibility seems a foveon like sensor if you want to go beyond that. And that if we do not look into the issue of corner sharpness with FF at wide angle lenses. I see with my 5D and the 24-70 already a lot of unpleasant corner issues at architecture or landscape shots where I am forced to take at least f 11 to get an acceptable performance.
Which brings me to the question if the next generation of pro SLR cameras are really that better resolution wise. They can bring better DR, better ergonomics and new features like dust remove or live view, but my "fear" is, that if you want top resolution with a decent corner sharpness you have to go the MFDB route - no way for me as I like to shoot outdoors in rough climate and my bread and butter jobs are at dim light (churches).

Christian

Actually, to get equal read noise per unit of area, it could double (in ADUs) or halve (in electrons). Read noise does not add, like photon counts do (shot noise doesn't add, either, it has the same square root relationship). It's only the photon *count* that adds linearly.

John,
Thanks for mentioning that. So in order to keep total image noise the same with 4x the pixel count on the same size sensor, each of the smaller pixels of half the pixel pitch, should have half the read noise. Right?

Would I be right in thinking that photonic shot noise should remain the same, irrespective of pixel count, provided that the same number of photons are reaching the same area of sensor, or does photonic shot noise for the whole image double in this situation of 4x the pixel count?

Probably not. The star performer is currently the 1D3. A 1Ds3 would not be a competitor to the 1Ds2 (which would become a discontinued model) but the 1D3.

This seems to me a misconception. Diffraction is a property of the lens. Until we succeed in producing matamaterials with a negative refractive index, suitable for camera lens construction, there's not much we can do about it without resorting to cumbersome techniques. There is no disadvantage in having a sensor which can outresolve a lens at a particular f stop. You won't get worse resolution at f16 because your sensor has more pixels than are needed.

There has always been a trade-off between getting maximum resolution and getting the desired DoF. It stretches across all formats. Large format photographers try to get around this by using tilt lenses. You can also use tilt and shift lenses with Canon 35mm DSLRs, but the TS-E 24mm seriously needs to be upgraded in my view.

There's also software currently available which can successfully merge a number of shots which have been bracketed with respect to focussing. This seems to me a process which could have great potential. Imagine, you have your 50mp DSLR on a sturdy tripod. You are using Canon's latest, highest resolving prime lens which is as close to being diffraction limited at f4 as most current lenses are at f8 (ie. double the resolution at the same MTF). The camera supports focus bracketing at up to 10 frames per second. You focus on the furthest point you want in focus, press a button; focus on the nearest point you want in focus, press a button; dial in the number of intermediate focussing points you need (or maybe the camera can work that out); press the shutter and voila!.

Hi,

To add to BJL's



and to Ray's



views I'd like to remind (although already discussed in earlier threads) of the possibility of getting along without AA-filter in the case of a sensor with high enough resolution. Personally I like 1Ds very much, it gives very sharp RAW-files because of a relatively weak AA-filter. The downside of it is of course the moire issues. Even Canon's WA-zoomlenses at f14 are sharp enough to introduce moire in 1Ds on certain subjects like scenes with far away office buildings with finely detailed vertical/horizontal structures. We have also seen moire issues in similar subjects when using 30+ megapixel MFDBs with sharp lenses.

In this sense I think we are not yet very near to the true limits of usable sensor resolution. Different topic is if the file sizes, noise characteristics and manufacturing costs etc are in reasonable relation to the potential IQ gains. Also for the added resolution to be fully exploitable the user has to be ever more careful in his/her technical craft. Except for the corner performance of many WA-lenses I think there's still quite a bit of room for added sensor resolution that can be put to use. Especially when making large prints it has always looked better to have camera-creted pixels (no matter how diffraction limited) than to have interpolated pixels.

Cheers,
Juicy

Right, relative to signal.



Shot noise is a function of quantum efficiency and exposure, only, per unit of area. Pixel-wise, it doubles (relative to signal) when fixing exposure and QE, but quadrupling the pixels per unit of area.

Okay! To get back to the hypothetical 50mp FF 35mm sensor. If I've understood you, such a sensor would have double the shot noise (over-all) as well as higher read noise compared with the 5D, provided we're talking about the same technology, same quantum efficiency of each pixel, same light gathering capacity per unit area, same read-out precision etc.

As technology advances in all related areas, it seems reasonable to me that such disadvantages will be overcome. The pixel pitch of such a 50mp camera would be around 4 microns, not much smaller than the pixel pitch of the latest 4/3rds DSLR from Olypus, the 10mp E-410 Volt with a pixel pitch of around 4.7 microns.

All Canon has to do to make such a 50mp camera worthwhile is design a few 35mm lenses of equal quality to the Zuiko lenses .