MP Equiv to Film Sq Inches: Why so Non-Linear?

[stevegust]

Why is it that 39 MP roughly equals 4x5 sheet film (roughly 20 square inches), yet 16.7MP only equals 645 film (roughly 3.87 square inches).  The relative difference in film area is over 5X, whereas the megapixel differences isn't even 2.5X.  Larger format lenses resolve slightly less lines per mm than smaller format lenses, but with color film, I thought the final difference in resolving power for film systems was quite similar....taken linearly, at 2MP per square inch of 4x5 film, 3MP would equal 35 mm film, but that isn't quite true by most people's results....

by extrapolation, will 80 MP equal 8x10 color sheet film if the digital sensor is large enough (i.e. 3x4 or 4x5)?

[normkoren]

Although the measurements in this report are not extremely precise (in terms of MTF, resolving power, etc.) the general trend you noted is quite correct. The issue is that real-world photography is not the same as ideal photography. Most large format lenses are sharpest around f/11. Sharpness is lost to diffraction as smaller apertures (larger f-numbers). As you go to larger formats, depth of field at f/11 tends to become hideously narrow, so most of the time (at least if you want DOF) you need to stop down well beyond f/11, which puts you into the diffraction-limited realm. It seems quite remarkable, but if you need extreme DOF, you become highly diffraction limited, and it almost doesn't matter what format you use. I explain this on

http://www.normankoren.com/Tutorials/MTF6....DOF_diffraction

Warning: This page his highly mathemtical. There are some other reasons why total resolution for film cameras doesn't scale linearly with film size. One is film flatness. It's not perfect for large formats; it can very with temperature and humidity. It's another reason that most large format photographers stop down beyond f/11. Another factor is atmospheric disturbances, especially for scenes at a distance.

I have seen a very few huge enlargements from 8x10 film, evidently taken under ideal conditions at optimum aperture, that are truly stunning, well beyond what you could do with any digital back. It's very hard to achive such quality under real-world everyday conditions.

[Anon E. Mouse]

It is not linear because it does not need to be. To make an 8x10 print, you are not magnifiying a 35mm negative the same amount you are a 4x5 negative. The 4x5 negative will give you a heck of a better print even though in absolute terms it is not "performing" at the level of 35mm. Since sharpness is a relative term defined in relation to the human visual system, 4x5 cameras can be stopped down to very small apertures and still provide sharp images.

And this is something I notice about the amateur community. They speak in absolute terms. To them everything is black and white. Professionals speak in relative terms and things are shades of grey. And here is the rub. While Micheal's test is a very good attempt for amateurs, it falls short on actually making any definate statements on which is "best" or even "better." Micheal alluded to the problem which is basically how an image looks - an extremely high-resolution imaging system does not always produce visually pleasing pictures. Imaging is really a systemic issue which is not dealt with at all. So while I found it interesting, I also found it hard to make any definate conclusions about the results because there were too many variables and unknowns.

But having worked 5 years in the production of digital cameras for a major Japanese manufacturer, I take great exception to Micheal's prejudice against optical and imaging theory. If it is good enough to produce the toys he likes using, then it is good enough to describe them. The problem again is a systemic approach. Just because he does not understand it nor do many people who use it in the silly film/digital debates, does not mean it does not work.

[Anon E. Mouse]

BTW, I looked at the same pictures as you and I thought the medium-format film resolved more than the 16 MP DSLR. Check the little loops on the right of the frame around the 1. The DSLR had better contrast, but where that comes from cannot be determined. But Micheal's test is really not very good at showing resolving power. It really is more about sharpness which is a much trickier affair to judge.

[michael]

"I take great exception to Micheal's prejudice against optical and imaging theory"

Well, there's an interesting assumption. A totally unfounded one, but interesting nevertheless.

A couple of comments, if I may.

I am a photographer. Not a scientist. I know that's no excuse. Bill Atkinson is both a photographer and a scientist.

But speaking for myself only, I have found remarkably little correlation in photography between theory, test results, and real world image quality. Our own little experiment reported on in Measuring Megabytes shows the pitfalls of even an attempt by three experienced photographers (one of whom applied scientific rigor to the process) can encounter.

On another note; I was a consultant to DXO Labs in the development of their DXO Analyser system. This is now in use by Chasseur D'Image, Popular Photography and a number of camera makers. I published camera tests here for a year or so using the system, but in the end stopped, because I found that difference in cameras reported by the tests didn't bear a whole lot of relationship to real world images. The graphs and charts made the whole thing look impressive, but I lost faith in them and couldn't share them with readers when my photographs from the field showed that what sometimes appeared like marked differences were not, while problems that I saw in image quality wern't reported by the test adequately.

And, with all due respect to my friend Norman Koren, whom I mention because he is in this thread, I think that while he offers with Imatest probably the finest testing system that I've yet seen outside of an optical lab, again, test results don't always coorelate to what is seen in practical use.

I'm not a Luddite when it comes to the scientific method. (I have some 7-8 US and international patents in my name, so I do have some experience in this area). I'm simply an unrepentant emperisist though. By this I mean that after doing photography, writing about it and teaching it professionally for some 30 years, I trust my eyes above anything else.

MichaeL

[paulbk]

“..an unrepentant emperisist"

Hear hear! All art is empirical. Scientific deconstruction of a image, while it may be valid, says little about the photograph. A human being deconstructed is a bucket of mud. The whole is greater than the sum of its parts.

[Kenneth Sky]

The proof is in the pudding - in this case the prints. What this forum needs to set up is a thread that can only be contributed to by those who have received the data disc, processed  and printed the pictures and judged them. The rest of us can only be voyeurs because talk is cheap (especially on the internet). But seriously, we need to await the distribution and evaluation of the data before making critical comments.

[Anon E. Mouse]

"Well, there's an interesting assumption. A totally unfounded one, but interesting nevertheless."

Well Micheal, you said at the end of your essay not to trust the theory. And this is not the first time you have made the comment. If the science and the results contradict each other there must be a problem in understanding the science (or the result). I also have a long career in photography and am well versed in the science. The science works. The problem is folks don't understand the complexity. They will take a single number and hold it up as some kind of truth rather than one piece of a puzzle that comes to form the final image. You actually recognized some of these problem in your essay.

Having just retired from the camera manufacturing industry, I find it ironic that folks think the science used to make the cameras is faulty, but demand perfection from their gear. That means the cameras, lenses, and other gadgets work coz of dumb luck - we are not talking about a new industry or science here. You will find the industry has many dedicated individuals working hard on this complex problem and doing it with more than a hit or miss approach. That is why your test does show how good cameras actually are. I think one thing we can agree on is all those cameras (and scanner) work very well. And inspite the range of manufacturers of cameras, lenses, and sensors, they, at least from what I saw in your essay, perform remarkably similarly. I certainly am not surprised the lenses test very close. MTF is great and VERY sensitive, but it is not a pass or fail test. Actually it is more of a diognostic test to make sure everything is tickey-boo. You may be surprised at how bad an MTF curve can be and still get good images from a lens. But I digress.

And yes, manufacturers do test their cameras in "real-world" situations at the end to make sure the thing works. Number tell so much and ultimately photography is subjective. (Lemons are rare and usually traced to a human error.) And the science is used to understand what is right or wrong with the final product and (hopefully) fix it.

Whether you take a good picture with the camera is up to the photographer. What constitutes a good picture is much harder. I know of one made with a bad lens (by modern standards), underexposed, on a bad emulsion, and not that sharp, but "behind the Gare St-Lazare" by Cartier-Bresson is still a good picture.

Now as far as the normal technical debates that rage on the internet, well lets just say a little information can be more dangerous than none at all.

But I have a feeling we will not agree on this topic.

[Ben Rubinstein]

That is an interesting point actually, maybe you could write an article explaining how the science and real world can and do differ?

[michael]

If the science and the results contradict each other there must be a problem in understanding the science (or the result).

I'll make a single observation in response.

Just as with high-end audio, where measurements do not necessarily correlate with what is heard, so too in high-end imaging.

The answer may be as simple as that we are either measuring the wrong things, or we do not know which things to measure, or how, to accurately represent the differences that we are seeing.

Michael

[normkoren]

Since I'm about as deeply involved in actual measurements as anybody, it's time to pipe in.

Scientific measurements are indeed valid and correlate well with visual image quality. But that said, there are a number of caveats.

First of all, image quality is the result of several factors, including MTF (sharpness, which is expressed as a curve, not a simple number), noise (which has spectral properties), tonal response curve, dynamic range, and color accuracy (pleasing colors are not necessarily accurate). These factors are summarized on
http://www.imatest.com/docs/iqf.html
And the image quality you perceive in a print is strongly dependent on post-processing. Imatest attempts to measure the quality of the raw material you have to work with, though it is affected by processing in the raw converter.

I've resisted the temptation to reduce Imatest results to a simple number. (I make an imperfect attempt to do so with Shannon information capacity, but signal processing (noise reduction and sharpening) makes it less than reliable.) Instead I provide users with tools they can use to arrive at their own conclusions. I expect that as time goes on, readers will refine their understanding of what constitutes excellent image quality.

I'm very familiar with the situation in high-end audio. Put simply, total harmonic distortion (which generated a marketing race in the 1970s) was a BAD measurement-- it didn't correlate with sound quality. I'll spare you the details here, but I've written about it, somewhat technically, in
http://www.normankoren.com/Audio/FeedbackFidelity.html

The audio issue correlates in a curious way with highlight burnout in (linear) image sensors. Long story, but the bottom line is that photographic measurements work a lot better than audio measuremets, but you can get in trouble when you try to reduce complex results (like MTF curves, noise spectra, and tonal response) to one or two simple numbers (even MTF50, which I use). As we become familiar with photographic measurements, which are new to many photographers, we should start to have a better idea how to interpret the numbers.

And by the way, if you don't plan to print larger than letter-size or A4, all the cameras tested are totally superb. They are superb at 13x19 inches as well. My EOS-20D, which is below the level of the cameras in this test, produces gorgeous 13x19 prints, far beyond anything I've ever seen with 35mm. The results in this test don't really apply to prints under about 16x24 inches.

[Gary Ferguson]

And by the way, if you don't plan to print larger than letter-size or A4, all the cameras tested are totally superb. They are superb at 13x19 inches as well.... The results in this test don't really apply to prints under about 16x24 inches.

Good point, and one that deserves repeating. But let me ask you a question.

In the days of film I used 35mm, medium format, and 4x5. Up to about a 10x8 print I couldn't normally see significant quality differences between these different formats. The key word being "significant". Because, using the very best lenses and most careful technique, I could still see some difference. However with digital capture and ink jet printing this relationship no longer seems to apply.

I now use a Canon 5D, a Canon 1Ds MkII, and a Phase One P25. Furthermore, I'll often use the P25 on a sliding carriage and stitch two frames together, giving a massive 40MP image. I agree with your point that for A4 or A3 all the cameras tested are superb, but I'm not sure your statement goes far enough. If I print to A4, or in some cases A3, using my range of cameras, say from 10-40MP, then I no longer see "no significant difference", I now see "absolutely no difference at all".

In other words, in the days of film a photographer printing to say 10x8 may have been satisfied with a 35mm negative, but could still have harboured a regret that they weren't using medium format for that extra fraction of quality. But in the digital world, with inkjets that simply discard pixels above a certain resolution limit, is there any point at all in having more pixels unless you're printing bigger and bigger?

[normkoren]

But in the digital world, with inkjets that simply discard pixels above a certain resolution limit, is there any point at all in having more pixels unless you're printing bigger and bigger?
[a href=\"index.php?act=findpost&pid=65923\"][{POST_SNAPBACK}][/a]

Hi Gary,

Since I'm using an 8.3 megapixel EOS-20D, which has fewer pixels and a smaller sensor than any of the cameras in the test, I couldn't make that statement myself. But now that you've made it, I'm happy to concur. 8x11 prints from the 20D are incredibly fine; it's hard to imagine better. So yes, there's no point at all in having more pixels (read, higher cost) unless you plan to print considerably larger. I suspect all the participants have Epson 7800 or 9800 printers. By the way, a friend from HP tells me they're coming out with some large format printers later this year that will give Epson a real run for it's money. I can hardly wait. I love big, fine prints.

[mtomalty]

I can hardly wait. I love big, fine prints.

I guess,then,  you'll be looking for a new camera   :>))

Mark

[Jack Flesher]

The results in this test don't really apply to prints under about 16x24 inches.

Amen!

And IMO this points to the core of the problem -- you have hoards of people commenting on the files that have never printed larger than A3.  After you have printed large for a bit, you develop an "eye" for what qualities will matter and what won't in a larger print.  I continue to be amazed when I find things in my large prints I never noticed onscreen, yet on careful review they were in the file all along.

[rmaes]

And then there is the whole (scientific) territory of RGB to CMYK transformation, and the representation of all these 10-40MP pixels through a stochastic dittering process. Many megabytes can be lost in this process.

If i want Big, Fine prints and get ultimate print quality out of the captured image. Shouldn't i rather opt for an RGB printing process (e.g. Lambda Durst) ?

[BJL]

Why is it that 39 MP roughly equals 4x5 sheet film (roughly 20 square inches), yet 16.7MP only equals 645 film (roughly 3.87 square inches).  The relative difference in film area is over 5X, whereas the megapixel differences isn't even 2.5X.
[a href=\"index.php?act=findpost&pid=65722\"][{POST_SNAPBACK}][/a]

Norm Koren and others have offered fine answers, but let me throw up one point: decreasing absolute len resolution (lp/mm).

As you move up in format size and focal length, lens resolution in lp/mm typically decreases, so doubling linear format size (quadrupling image area) probably gives significantly less than double the "line pairs per picture height" of resolution, or in the new digital currency, it less than doubles "pixels per picture height", meaning significantly less than a quadrupling of the "pixels worth of detail".

Photodo made a similar observation in a resolution comparison of 35mm, MF and LF systems using Velvia, concluding that in 35mm resolution was largely film limited, while in LF resolution was largely lens limited.


P. S. Another possible factor is scanning rsolution: Michael mentions in another thread that the 645 was scanned at about 4500dpi while the 4x5 was scanned at 2450dpi, so that is a potential factor reducing the "lp/mm" of the larger format scans. However, 2450dpi is close to the 50% MTF of the film (Velvia), so this might not make much difference.

[Ray]

As you move up in format size and focal length, lens resolution in lp/mm typically decreases, so doubling linear format size (quadrupling image area) probably gives significantly less than double the "line pairs per picture height" of resolution, or in the new digital currency, it less than doubles "pixels per picture height", meaning significantly less than a quadrupling of the "pixels worth of detail".

That's a good point, JBL, which could be summarised as the 'law of diminishing returns' at work. We now have cameras that have 4x the pixel count of Canon's first DSLR, the D30. I checked the dpreview site for a comparison of lines per picture height for the D30 and the 5D. The table below is not exactly analagous to the situation you refer to above. The 5D has 4x the pixel count of the D30 but only 2.5x the area. Nevertheless, I am surprised  the 5D really does seem to deliver twice the 'lines per picture height' of the D30. However, I wonder if the 5D  resolution per picture height is an extrapolation of resolution at the centre of the lens and would only apply in practice using a lens with an unusually flat MTF response.

[attachment=608:attachment]

[bob mccarthy]

As you move up in format size and focal length, lens resolution in lp/mm typically decreases, so doubling linear format size (quadrupling image area) probably gives significantly less than double the "line pairs per picture height" of resolution, or in the new digital currency, it less than doubles "pixels per picture height", meaning significantly less than a quadrupling of the "pixels worth of detail".

I don't know if that holds water anymore. The modern LF lens is darn good today. I'm seeing resolutions of 60-80 LP-MM with the newest generation and the resolution holds up quite wel, nearly to the edge of the image circle in the Super Symmars.

http://www.hevanet.com/cperez/testing.html

The reduction in image magnification is what makes the LF film camera interesting. Much greater apparent DOF before defraction kicks in.

(wishful thinking) a 4x5 chip with big wells would be the ultimate capture device. Only a government could afford it though.

Bob

[BJL]

I don't know if that holds water anymore. The modern LF lens is darn good today. I'm seeing resolutions of 60-80 LP-MM with the newest generation and the resolution holds up quite wel, nearly to the edge of the image circle in the Super Symmars.

The reduction in image magnification is what makes the LF film camera interesting. Much greater apparent DOF before defraction kicks in.
[{POST_SNAPBACK}][/a]

I disagree on both points: smaller format lenses maintain a very big resolution advantage over even the latest lenses for 4"x5" format, measured by MTF at equal lp/mm, and the trade-offs possible between DOF and diffraction are the same for any format. Only the use of view camera motions given any advantage to a larger format on apparent DOF when diffraction effects are equal, and view camera motions are coming to 35mm format digital now.


I) The Super-Symars you cite are still not even close to good 35mm format lenses for "absolute resolution", MTF at equal lp/mm. (They probably are sharper in line pairs per pictuer height, and thus for lp/mm on prints of equal size.)

Compare the MTF curves for two normal primes, the Super-Symmar XL 210mm f/5.6 at
[a href=\"http://www.schneideroptics.com/photography/large_format_lenses/super-symmar_xl/pdf/super-symmar_xl_56_210.pdf]http://www.schneideroptics.com/photography...r_xl_56_210.pdf[/url]
and the Canon 50/1.8 II at
http://www.usa.canon.com/consumer/controll...52&modelid=7306
I see the Canon having better MTF at 30lp/mm than the Super-Symar at the lower 20lp/mm, and the Canon at 10lp/mm is way ahead of the Super-Symar at 10lp/mm or even at 5lp/mm. And that is comparing to a very inexpensive non-"L" lens! Comparing instead to the Canon 50/1.4 (still not an "L" lens, but pricier) gives similar results.

Going to an even smaller format, the inexpensive Olympus 4/3 format 35mm f/3.5 macro lens http://www.olympusamerica.com/e1/sys_lens_35mm.asp
does even better fr lp/mm of resolution. Its MTF at 60lp/mm and wide open f/3.5 is far better than the Super-Symmr at 20lp/mm and any aperture, and the 35/3.5 has MTF at 20lp/mm as good or better than the Super-Symar at even 5lp/mm.
Even a mid-priced small format 4x zoom lens like the Olympus 14-54/2.8-3.5
http://www.olympusamerica.com/e1/sys_lens_14mm.asp
has better MTF at 60lp/mm (wide open or f/8) than the Super-Symar at 20lp/mm, and the 14-54 at 20lp/mm trounces the Super-Symar at 20lp/mm and 10lp/mm, and outdoes even the 5lp/mm Super-Symar performance.

At the risk of being "too theoretical", any resolution improvement in a larger format can carry over to a smaller format, and smaller format makers like Canon have a lot more money and resources, so are certainly capable of making such improvements if they choose. Downsizing a LF lens design in all dimensions by a factor of four shrinks the focal length by the right factor to give the same angular FOV, and shrinks the size of all "design" abberrations by the same factor, essentially giving four times the resolution (same MTF ast fout times the lp/mm). This downsizing  gives equal aperture ratio, as was used in the tests (f/11 for all). Arguably (see below) the smaller format woud mor often use a lower apertrure ratio, worsenign abberrations.


II) On the second point about DOF and diffraction: the possible balances of DOF and diffraction effects on print sharpness are independent of format, once you level the playing field by viewing prints of the same size from the same distance. In fact, they are each determined solely by the effective aperture diameter (focal length divided by aperture ratio), so you get the same DOF and diffraction results in different formats by adjusting aperture ratio in proportion to focal length.

Why? Famously, if a larger format and longer focal length are used with the same aperture ratio, the larger format gives less DOF. In detail, the sizes of the circles of confusion on the film or sensor increase with the square of focal length, but then degree of enlargement is reduced in proportion to focal length, so the CoC sizes on prints are larger in proportion to focal length.
To get equal perceived DOF, this must be compensated for by increasing the aperture ratio in proportion to the focal length (that is, using the same effective aperture diameter).
Then the diffraction spot size on the film/sensor is determined by aperture ratio alone, independent of focal length, so once you choose apertures to get equal DOF, the diffraction spot size on film/sensor is larger in proportion to focal length.
The smaller degree of enlargement needed with the larger format just cancels this out, so the diffraction spot sizes (and circles of confusion) are the same size on same sized prints, from any format.


P. S. Large format enthusiasts should probably stick mostly to the legitimate claims of
1) greater print resolution in lp/mm on equal sized prints, after the lower degree of enlargment needed
and
2) finer tonal gradations/greater dynamic range
to explain this preference. Those two have always been and probably always will be the main real reasons to sometimes choose a larger format than most others photographers use.