An argument against consumer full-sized sensors

[b.e.wilson]

[font color=\'#000000\']Oops, never mind. I see RS Foto alluded to this in a reply on the first page of the "Nikon Strategy, Cost of Imaging Chips" thread, which is still active.[/font]

[Rainer SLP]

[font color=\'#000000\']Hi PDW,

It will not help. The Lensmakers will always have the problem of the edges if you adjust the image circle to the recording Rectangle or Square.

Only if you really make your image circle bigger and use the best portion than you will get excellent results with your lenses.

Sure they can already design better lenses, but, yes here comes the famous but, who will be willing to pay for them. Look at all the discussions about the prices of the CANON L lenses?

As long as nobody is willed to pay for quality they will keep making the lenses " As Good as Necessary and Not As Good as Possible "

Do you know what the 1200mm f5.6 from Canon costs? around US $ 85,000.00 and this is an example of a good lens.

OK you can get an 800mm f11 mirror lens already for US 500.00 but is it the same?[/font]

[font color=\'#000000\']The only confimed owner of the 1200mm f5.6 Canon lens that I know of is Sports Illustrated magazine. They have one and ship it around the country to their staff and contract photographers who might need it for a given event.

Apparently it costs about $400 to ship it each way by Fedex.

Michael[/font]

[b.e.wilson]

[font color=\'#000000\']PDW, that's my point. The lenses that match these full-sized digital sensors exists, they just cost five times what a comsumer lens costs. It almost seems you're advocating bumping the L line down to consumers, which we know is very unlikely with current lens-making technology.

Consumers have lenses already, the lenses that sell the most, and they aren't L lenses. What I argue is that an APS-sized sensor will give better results with one of these lenses than a full-sized sensor will, and that end quality will dictate the consumer digital body market until the lenses change.[/font]

[b.e.wilson]

[font color=\'#000000\']

Note that lack of sharpness due to DOF is worse than indicated by the traditional DOF guidelines, which only guarantee a distinctly lower degree of sharpness than is given by decent consumer films and by good consumer zoom lenses at the focal plane (circle of confusion of 0.025mm or 0.03mm for 35mm film, bigger than diffraction spot size for anything down to f/22, and equivalent to only about 30 or 40 lp/mm).

Recently I did a bit of research on the history of Depth of Field, and found that the DOF markings on lenses, and the 0.032 mm circle of confusion that corresponds to them, were defined a long time back, when TMY was a new and superior film. For the circle of confusion to match the accutance of modern slide films like Velvia and Provia100F, you must use a cof figure more like 0.005 mm and accept a restrictingly narrow DOF range.

But the COF is defined in term of the final print 3 lp/mm (100 lp/inch), the generally-accepted limit of visible distinction, not the film sharpness. Oddly, the 0.032 mm value assumes perfect fidelity in the enlargment, which isn't true even of a contact print.

So here is a very unscientific (but based on the proper equations, observations, and measurements) rundown of 50% MTF line pairs/mm seen on an 8x10 print from 35mm film:

35mm film COF of 0.032mm: 22.5 lp/mm on film, 2.5 lp/mm on paper.

f/22 diffraction: 34 lp/mm on film, 3.9 lp/mm on paper

Lens accuity (green light, central zone only): 45 lp/mm film, 5.2 paper

Film accuity (green light, low contrast, ISO 100 slide): 40 lp/mm film, 4.6 lp/mm paper

Film bulge (roll film): 70 lp/mm film, 8.5 lp/mm paper

Total expected sharpness based on these criteria only: 8 lp/mm film, 0.9 lp/mm paper.

Now add camera shake, lens aberrations and flare, enlarger shake, focus problems, dye diffusion in the print, and you'll quickly arrive at the conclusion that sharp prints need a lot more sharpness in the film than what the 0.032mm COF predicts. Is it any wonder the lens makers dropped those near meaningless DOF lines from the lenses?

Sure, this is less than a back of an envelope calculation (based on my gross simplification the very complex multiwavelength MTF curves of each 'light processing' step as a single value), but it sould be good to give you an idea of how each step of the photogrpahic process relates to overall sharpenss.

Anyway, to get back on topic, for the low-cost lenses, aberrations and geometry problems start to compete with the more unavoidable contributors to unsharpness, but they can be minimized by ignoring the corners. Then there is just everything else messing up our sharp images.

By the way, I did the calculations using a small database applet I wrote for the Pocket PC: http://chem.dynu.com/photo/photoassist.asp[/font]

[Ray]

What I'm having difficulty in understanding is the advantages of the larger format should it ever be possible to produce 3 or 4 micron photodetectors that have a performance on a par with, say, the current 8.8 micron 'pixels' of the 1Ds. Where are the bottlenecks? Are the higher quality P&S fixed lenses REALLY diffraction limited at F4? Is it possible to produce an 8mm lens that is diffraction limited at, say, F2. I mean, a lens that is diffraction limited at F2 should have a resolving power of 500 or more lp/mm!! Is such performance ever going to be possible at an affordable price? What is possible at the present time? Where's the cutting edge? Where are the current limits?

[BJL]

Is it possible to produce an 8mm lens that is diffraction limited at, say, F2. I mean, a lens that is diffraction limited at F2 should have a resolving power of 500 or more lp/mm!! Is such performance ever going to be possible at an affordable price? What is possible at the present time? Where's the cutting edge? Where are the current limits?

I am very interested in this question too, but I am a little pessimistic about tiny sensors.


1) Total detail, in lp per picture height.
When comparing different image or sensor sizes, it is mostly total detail in the sense of lp per picture height that counts: if f/2 gives diffraction limitation to 500lp/mm in a P&S size sensor of about 6mm height, that is 3000lp per picture height, the same total detail as 125lp/mm in 35mm format, corresponding to diffration limitation at f/8, but the P&S requires one quarter of the pixel spacing. (And as I have mentioned before, the DOF is the same for those two hypothetical lenses at those respective f-stops.)


2) Minimum pixel spacing for exposure latitude and "digital grain".
Digital sensors are already reaching about 50% of theoretical maximum efficiency in detecting light for a given pixel spacing, so only about one more f-stop of improvement in signal to noise ratio ("digital grain") and dynamic range is possible at a given pixel size. So it might be impossible to ever go below about 5 or 6 micron pixels without having exposure latitude less than slide film (and about twice as big to match negative film's greater latitude.)


3) Scaling of lens aberrations with image size.
For the same angular field of view, lenses for smaller image or sensor sizes with proportionately smaller focal lengths and smaller image circles probably can control aberrations down to smaller absolute sizes (lp/mm), but probably only in proportion to overall size reduction, and if everything scales with image size, the limit for a lens of the same angular field of view would be the same at the same aperture ratio; perhaps that is where Norm Koren gets his universal f/4 guideline.

Optimistically, that f/4 might assume traditional spherical lens surface shapes. If improvements like better aspherical lens designs can reduce aberrations to match diffraction at f/2, the same designs could probably be used to get to f/2 for larger format lenses too; but at higher cost as always.

Improving the aberration limited f-stop with the same minimum pixel size pushes the minimum "full quality" image size upwards.


P.S. If we accept f/4 and about 5micron pixel spacing as lower limits then the lower size limit for a format limited only by diffraction, unavoidable lens aberrations and film-like exposure latitude is just a bit smaller than 35mm format, probably in the range from the proposed 4/3" format (13.5mmx18mm) up to the biggest of the current "APS DSLR" formats (16mmx24mm).

[b.e.wilson]

[font color=\'#000000\']How would a camera like the Canon D1s perform with consumer lenses?

Clearly the thing is designed and priced to be used with L lenses. But how would it perform with the popular 28-105 or 75-300IS lens? Both those lenses, while good, definitely have a sweet spot in the center of the image circle, especially when opened up.

With a small sensor most of the image is in the sweet spot. If a large sensor were used, the corners would be less sharp, and geometric distortions more evident.

By sticking to a smaller sensor, the problems with the less-accurate lenses are minimized, clearly an advantage to non-pro lens users. The trade off is less pixel resolution everywhere in the image for better image quality at the corners and better geometry.

Even if large sensors were a buck a piece, I still think manufacturers will opt for the smaller sensors, unless unreasoned market forces compel them to change.

And I beg your pardon is someone has already realized and stated this. I did a search but found nothing.[/font]

[PDW]

[font color=\'#000000\']Or, the camera manufacturers could just design better lenses.  It seems a bit stupid to always carry around lenses with extra covereage.  Also, if the ff sensor has enough resolution, it's possible to just crop and end up with the same result as an APS sized sensor.

Aside from physical size, I can see no downside to full-frame sensors.[/font]

[Marshal]

[font color=\'#000000\']I've always been curious about how many people own that 1200mm f5.6 lens at its $85k price.[/font]

[PDW]

[font color=\'#000000\']I'm no lens designer though I'm not completely ignorant on the topic.  It is possible to make a lens that is sharp edge-to-edge though it probably is easier to make a lens with more coverage and only use the center.  However, more coverage costs more to produce and it means a bigger, heavier lens and more coverage means hard to design lenses (fast, wide, big zooms) will be even harder to design.

Up until a few months ago, the most resolution you could get (in a 35mm type body) was using the best film and the film was likely to be the limiting factor.  Now, with the 1Ds, the lenses seem to be the limiting factor but you're still getting more resolution than before.

I guess I don't see this as much of a problem.  Buy good lenses and you'll have enough resolution.  As you state, good lenses are already made though there's a lot of them that are cheaper than your 1200/5.6 example.  Imagine that lens if it had to cover more area.[/font]

[PDW]

[font color=\'#000000\']Actually, what I'm advocating is Canon L or Nikon ED quality lenses at smaller max apertures and therefore cheaper.  More lenses like the Canon70-200/4 for example.  I'm also saying that the manufacturers have already done this to a degree.  The new Nikon 24-85G is an ED lens and is supposed to be excellent.  The new 12-24 is also an ED lens and was designed to cover the APS sized sensors (though I have no idea how good it is and how much it covers).  Alternatively, use primes like they did in the old days.

There's also diffraction limits that mean a larger sensor will give better images all other things being equal.[/font]

[BJL]

[font color=\'#000000\']There's also diffraction limits that mean a larger sensor will give better images all other things being equal.[/font]

[font color=\'#000000\']I am tempted to respond that
"There's also depth of field limits that mean a SMALLER sensor will give better images ..."
Or in other words, "all other things are NOT equal".

In close portraits, the DOF limit is so severe that getting the full sharpness at the focal plane that good 35mm film can record means accepting noticably less sharpess even a few cm away: examine the ears on a portrait that is very sharp at the eyes.

In a sense diffraction and DOF cancel out with change in sensor size or film format: if you change to a smaller sensor, the change to a larger f-stop needed to get the same diffraction spot size relative to overall image size (so smaller on the sensor) takes you back to the same depth of field.

So I have a rather vague question: in choice of sensor or film format, how should one balance improving sharpness in the focal plane against maintaining adequate sharpness over all of a three dimensional subject which requires good sharpess over at least some range of distances? This is I believe part of the reason why a lot of professional and serious amateur photography is done with the formerly "too small, low-grade, snap-shot only" 35mm format.

Note that lack of sharpness due to DOF is worse than indicated by the traditional DOF guidelines, which only guarantee a distinctly lower degree of sharpness than is given by decent consumer films and by good consumer zoom lenses at the focal plane (circle of confusion of 0.025mm or 0.03mm for 35mm film, bigger than diffraction spot size for anything down to f/22, and equivalent to only about 30 or 40 lp/mm).[/font]

[Hank]

[font color=\'#000000\']Interesting and enlightening discussion.  All a little academic for consumer P&S cameras, but cogent if the D30, D60, D100, et al, may be considered consumer DSLR's.  

Maybe I am being selfish, but I'll be happiest when I have it ALL in a DSLR:  I love the magnification effect of the smaller sensor when using long lenses for wildlife, but it would also be nice to move to a full-sized sensor when I need the extra wide angle coverage.  I buy the best glass that I can afford, then work within the limits of a particular lens.  If that means a compromise between focal length, f-stop and the cropping tool in Photoshop, so be it.  With the ever-larger files from DSLR's I think we will have the latitude to do more cropping in many cases.

Stir all that mush into my kettle of soup, and my ideal DSLR will have a CMOS 24x36mm sensor with features to allow me to "turn off" the outer fringes, effectively reducing its size and introducing a magnification factor.  Barring that, I'll take the full-sized CMOS sensor and apply my own magnification factor using the PS cropping tool.

Hank[/font]

[JCDoss]

Actually, what I'm advocating is Canon L or Nikon ED quality lenses at smaller max apertures and therefore cheaper.  More lenses like the Canon70-200/4 for example...

I would just LOVE it if Canon would release a series of L-zooms similar to their big three (16-35L, 24-70L, 70-200ISL) with an f/4 max aperture and with a wider range.  Even if they kept the same focal length range in each lens, imagine the savings in price and weight!  It would be even nicer if they could extend the range of the middle and long lenses to something like 24-100/4 and 100-300/4 with IS in the latter.  And if we ever needed speed, we could turn to fast primes, or alternatively (as sensors improve) higher ISO.

I also agree that having an "all-in-one" style full-frame sensor would be the best option, and I believe we will see sensors in the coming years with croppable magnifiers and aspect ratios.

[Ray]

BJL,
You seem to have been doing a bit of recent reading on the topic ::

I'm beginning to think top quality images will always remain the preserve of MF. As 35mm approaches the performance of MF, digital MF backs will eventually exceed the current performance of large, film based, field and panorama cameras, and there will probably always remain that huge price differential.