Re: Digital vs 35mm camera optics

From: Stephen H. Westin (westin*nospam_at_graphics.cornell.edu)
Date: 08/20/04 

Date: 20 Aug 2004 12:45:18 -0400

rmonagha@engr.smu.edu (Bob Monaghan) writes:

> well, let's see, y'all were the one originally citing 100-125 lpmm IIRC;

This is where your habit of not quoting gets to be a problem.
Three days ago, I sent the following message:

rmonagha@engr.smu.edu (Bob Monaghan) writes:
>
> > not even close ;-)
> >
> > steve stated:
> >
> > As far as resolution goes, film is frequently capable of 100-125 cy/mm.
> > This corresponds to 8-10 Mpixels. You can already buy 8Mp prosumer
> > cameras, so the performance of these lenses should be comparable to the
> > less expensive SLR lenses.
> > end-quote:
> >
> > since a cycle/mm is a line pair per mm, that's 2 pixels (one black, one
> > white) per cycle, or 200 to 250 pixels/mm. for 35mm film, 24mm x 36mm,
> > that's 4800 x 7200 pixels (using 100 cycles/mm to be conservative here),
> > or 34.56 megapixel equiv. Hardly 8 MP, but four times+ the resolution,
> > right?
> >
> > Use the 125 cy/mm value, and you have 6k * 9k or 54 megapixels equiv.
>
> Unfortunately, film isn't capable of that resolution in any normal
> situation. For a good ISO 100 color film such as Velvia, the MTF is
> down to 50% at just 50 lp/mm. The quoted "ultimate resolution" is at a
> contrast of 1600:1, which isn't achievable in any normal camera
> photographing a normal scene. Also, 35mm lens MTF seems to be reported
> at 20 and 40 lp/mm, with significant decline at the greater spatial
> frequency. So if we take, say, 60 as a conservatively high number,
> that would be 120 pixels/mm, or a sensel pitch of 8.3
> micrometers. Which is right in the ballpark of current digital SLR
> sensors. The result is about 12.4MP, which splits the difference of
> the two full-frame DSLR's on the market, the Canon EOS-1Ds at 11MP and
> the Kodak models at about 13.5MP.

So as far as I can see, the 100-125 figure is yours, and I was disputing it.

> I
> am simply pointing out that the pixel equiv. is vastly higher than the 8
> Megapixels equiv. claims being made

...but not by me...

> by simple math of what 125 lpmm means
> if we convert to pixels, 2 pixels per line and 35mm format size.

> Let's do it the other way. For mid speed 35mm film to be 8 Megapixel
> equiv., it would need to be circa 100 pixels per mm (as with full format 8
> MP digital cameras, obviously, right?). That means 50 lpmm MAXIMUM for
> 35mm film cameras. Now go check pop photo's historical database of lens
> resolution tests for Nikon, Canon, Minolta, Pentax, Zeiss, and so on. See
> how many lenses scored in the 70 lpmm and 80 lpmm range?

Again, I'm not really ready to take Popular Photography's figures at
face value. Choosing a suitable Zeiss lens, the Sonnar T* 180 f/2.8
at
<http://www.kyoceraimaging.com/lensdetail.asp?brandid=&parent=123000>,
we see that, wide open, the MTF at 40 lp/mm never reaches
0.3. Multiplied by a typical color film MTF of 50% at 50 lp/mm, we get
only 15% of the original contrast! Can you show me how that looks in a
real scene? At f:5.6, we can get up to 25% of the scene contrast, but
only in the center.

Note that this is an expensive lens under laboratory
conditions. Perfect focus, no vibration, etc. It only goes down from there.

> I have a page on lens manufacturing variations on my site at:
> http://medfmt.8k.com/third/variations.html - lots of example lens test
> reports at 70 lpmm, 80 lpmm, and even beyond - and this was with older
> film stocks from the 1960s onward too ;-) Most of these lenses are normal
> or 50 mm focal length lenses, often the cheapest lenses in these lens
> mfgers lines too. And careful testing of these same lenses with higher
> resolution films cf. http://medfmt.8k.com/mf/limits.html#100 concluded
> "The first thing which can be inferred from the tables is that the 100
> lines per mm level of image sharpness can be reached by many normal focal
> length lenses when the aperture is set at f/4 or f/5.6, and the "best"
> film is used." - from a 1978 study...
>
> So either all lens testing for the last 40 years is wrong, and lenses
> can't really beat 50 lpmm as an 8 Megapixel equiv. claim necessarily
> implies, or those claiming 8MP (or anything below about 32 MP) is better
> than (midspeed ISO) films are simply obviously wrong, again by simple
> math! ;-) The reason is the Nyquist limits and other issues with scanning
> film, IMHO, and the use of low resolution display monitors or 300 dpi
> (equiv. to a modest 5.6 lpmm on paper) prints to evaluate such "tests" ;-)

OK, then why isn't aliasing a problem in every image the Kodak SLR's take?
If there really is that much scene content beyond the Nyquist limit of
62.5 cycles per millimeter, shouldn't every image have horrible moire?

You are, I hope, aware of the thoughts of Erwin Puts, a decidedly
non-digital guy, on real-world resolution. If not, try
<http://www.imx.nl/photosite/technical/resolution/resolution.html> and
follow links.

For example, at <http://www.imx.nl/photosite/technical/highres.html>,
"The maximum figures quoted out of context for diffraction limited
lenses and high resolving powwr for micro-films are theoreetically
true no doubt about that. When translated into general photographic
practice, these figures need to be downgraded very much to allow for
optical aberrations, system resolution, contrast drop, picture
technique, camera vibration etc.

Even allowing for all these effects the goal of 100 lp/mm on the
negative is hard to reach. On the other hand, this study has shown
that the 100 lp/mm are the limit of what we need (or can use) for high
resolution imagery with 35mm equipment.

Most photographers do not care whether they are photographing with 20
or with 80 lp/mm, as they are interested in good content and set high
values on art and emotion as parameters for good a good picture.

No trouble at all. But fro[sic] those who want to know where the limits
are, it may be a sobering thought that the jump from 20 to 80 lp/mm,
while seemingly a small step, is very difficult to attain. And the
proposal that 600 lp/mm are viable on the negative is theoretically
and practically illusory."

His arguments seem convincing to me, and he documents his procedures.

<snip>

> You reject both Kodak's 24 MP estimate

Sorry, I missed the reference on that. Where did Kodak state that estimate?

> and Pop Photo's real world test
> report; yet Kodak has bet the corp. future on digital, so has no reason to
> push the superiority of film ;-), ditto Pop Photo and Imaging, as their
> new name suggests, the digital end is now the main component of their
> focus, articles, and advertising revenues. In the legal biz, we used to
> call these "statements against interest", and they are among the highest
> level of proof you can offer in a legal argument ;-)
>
> Similarly, it is relatively simple math that if the lens+camera+film is
> delivering 100-125 lpmm SYSTEM resolution,

Which it isn't, under most circumstances,

> the individual components must
> be rather higher.

But, of course, the film image has to go through another set of optics
to be viewed, so the resolution of the final image is generally lower
than what is captured on film.

> The eye can readily work down to and below 2% contrast
> levels,

At certain angular frequencies and illumination levels. Using my
well-lit office as an example, I have about 210 cd/m^2 luminance of a
white *** of paper on my desk. Using the contrast sensitivity
function from <http://webvision.med.utah.edu/KallSpatial.html> and the
chart at <http://www.pc.ibm.com/ww/healthycomputing/vdt13eyec.html> to
convert cd/m^2 to trolands, it seems that the top angular frequency
perceptible at 1% contrast is about 30 cycles per degree. By my
calculations, that comes out to about 0.35 millimeters per cycle, or
144 dpi at 2 foot viewing distance. That's about 2800 pixels across a
16x20" print. The end of the curve is for about 10% contrast, at 50
cycles per degree, which comes out to about 4700 pixels across a 16x20
print. So about 15 MP ought to do it. Notice how quickly the
sensitivity falls off (i.e. the required contrast increases) as the
angular frequency increases.

> and some scanners go below 5%, so why should we agree to only
> consider film images at some magic 50% contrast level or above? As zeiss'
> tests suggest, these high resolution results can be achieved with consumer
> films, using commercial processing labs, and high quality lenses ;-)

It's just way more complicated than that. The figures I gave above
were based on tests with pure sine-wave gratings; a real image is much
more complex, probably making the visual system less sensitive to high
frequencies.

Besides, you are assuming a lot of high-frequency content in the
scene, which generally doesn't happen. A step edge has a spatial
frequency that drops off as 1/f in magnitude; most of its energy is at
lower frequencies.

> re: digital cameras w/o anti-aliasing filters You can use a high end
> digital camera without an anti-aliasing filter, you just will get rather
> higher aliasing artifacts and other related problems.

Actually, Kodak has made the only high-end cameras that can be used
without an AA filter; the older ones had it removable, and the current
ones don't have one at all. Oh, and of course the Sigma/Foveon
cameras. What others are there?

As to "higher aliasing artifacts and other related problems", they
don't seem to occur in most images, which supports my assertion that
the high-frequency content you worry so much about might not be
that common.

> The poor sales of
> the Kodak camera series lacking AA filters is well known.

Or do you mean "the poor sales of the Kodak camera series with CMOS image sensors"?
Or perhaps "the poor sales of the camera series marketeed by Kodak"?

Anyway, I don't see how you opinion of their sales levels (Kodak might
have a different view) makes the cameras not exist, or reduces their
resolution capabilities, or makes them incapable of film-like
resolution.

> The poor market
> response to the Sigma SD9 camera is also related in part to its lack of an
> AA, among other issues, in the form of luminance aliasing. see comments at
> http://www.imaging-resource.com/PRODS/SD9/SD9A12.HTM - which also notes
> and explains some of the spurious user "tests" which report resolution
> values above Nyquist limits, and casts doubts on many scanner tests in
> general...

Again

> Why do I insist that scanning film is biasing results? Because it
> obviously is ;-) We know from the film maker's statements (Cf. Kodak),

Which I would like to see quoted and cited in context.

> independent lens maker's tests (cf. Zeiss tests),

Of which we have only the sketchiest of details, and no images
demonstrating that the measured resolution is visible.

> and independent magazine
> lab tests (and pop photo & imaging is one of the few USA mags with an
> actual megabuck photo test lab setup),

And when did you last see them say a new lens or film was low in
resolution?

> what the megapixel equiv. of film
> should be. The answers for mid-speed 35mm film are all in the 24 MP to 50+
> MP range, depending on the mid-speed ISO film tested.

And Erwin Puts says otherwise, testing in real-world conditions.

> If you are getting answers from scanned film that film is worse than 8 MP
> or even worse than 6 MP, then I have to believe the problem lies in the
> scanning of the film, not in the lab tests of the film mfger (again, the
> #3 digital camera maker too ;-) and other professional and rather better
> equipped and trained labs than the average USENET photo geek with a
> scanner ;-) I tend to give the mfger's reported specifications greater
> reliance than what I read on USENET too ;-)

What about the scanned print? Do you believe that flatbed scanners are also
incapable of their rated resolutions?

> My point about the use of a loupe is that film has considerable fine
> resolution detail which I don't see in similar digital imagery at similar
> enlargement factors.

"similar digital imagery" whose source or character you haven't seen fit
to reveal.

> Again, this is not surprising, because the film has
> higher resolution and no low pass filter in it (and probably 99.8% of all
> digital cameras have some sort of low pass filter in there for
> antialiasing etc.). I could do enlargements to 15X or larger, but I
> generally don't, largely due to costs and a lack of wall space and ceiling
> space for display (you know it is bad when you have stuff posted on the
> ceiling, right? ;-)

Let's see. 15X from 35mm would be about a 14x20" enlargement. It works from
a good digital, too.

> Having Leitz and Rollei slide projectors for 35mm and
> 6x6cm is one reason I shoot slides. Try projecting a digital slide to wall
> size, and see how it looks compared to a velvia 6x6cm chrome to the same
> size. That's why I'm sticking with film ;-)

Hmm. You were claiming that digital sensors are far below film
resolution, but now you seem to be comparing a 24x36mm digital sensor
with a 56x56mm slide. Are you backing off?

I do get the feeling that you don't bother looking at the Web sites
I point you to, but would rather not gain any new information. Am I
right? 

-- 
-Stephen H. Westin
Any information or opinions in this message are mine: they do not
represent the position of Cornell University or any of its sponsors.
Quantcast