« Which Came First, 'On' or 'Off'? | Main | Random Excellence: Animesh Ray »

Sunday, 30 September 2007


Feed You can follow this conversation by subscribing to the comment feed for this post.

From his blog entry, this is the MTF for the Ricoh GRD lens.

Dear Mike,

That's a nice representation. It has some real-world relevance.

You've said this is a very good lens, based on the MTF. Readers should note that going from 10 lp/mm (a 100 micron blur circle) to 30 lp/mm (a 35 micron blur circle) drops the average MTF from over 90% to under 70%. Assuming this lens performs similarly to others of that quality, at 60 lp/mm (15 micron blur circle) the MTF will be down around 30%, less than 1/3 of "normal" contrast.

That's still a pretty large blur circle. Any modern film or digital sensor can make use of considerably finer detail than that. Typically, at the point at which you max out the film (or sensor) that way, you're looking at MTF's down around 20% or less.

What's the import of this? That arguments about films' vs sensors' innate MTF's miss this point that lens performance is still a major determinant of image quality. No matter how ideal you think a sensor is, the lens system is going to dictate contrast rolloff, even in optimal situations. In less-than-optimal ones, other factors will degrade MTF's even faster.

This also relates to the uselessness of abstract discussions of diffraction limits. By the time you reach the diffraction limit of lenses, regardless of aperture or lens, you're normally dealing with extremely low MTF values. Those values are not a major component of overall image sharpness. Diffraction limits are not some dangerous, magical boundary between good and bad image quality that one risks falling off the edge of; they're simply one more gradual source of MTF degradation among many.

pax / Ctein

Erwin Puts explains how cropped sensors affect lens' performance:



I'll make a lateral statement about another quality that looks very good to me in this chart and isn't frequently explained. This lens displays a very flat behavior from center to near the corners. Others might peak at a very high point but then vary a lot through the image. I find this last profile to be very annoying, even if the lower points are quite good. Why? because we have digital sharpening now. With optical printing and enlarging it wasn't an issue, because there wasn't much you could do about it. But now, when a lens has a mediocre but very even resolution behavior from center to corner (ideally a horizontal flat line), you can easily sharpen all of it and improve the results immensely, and the picture will look natural. When you have a spike in the graph, your picture has a lot of different sharpness levels depending on the area in question, and it's a lot tougher to get an even-looking result with the sharpening tools. Obviously, if you shoot portraits it's not as bad an issue, but when you want even sharpness, it's quite a challenge.
In short, I love a lens that is extremely sharp everywhere, but between a lens that produces good even resolution levels and one that is outstanding at the center but varies a lot through the frame I'd rather get the "just good" one. It's a lot easier to work with.
That's why also I'd go for f/32 with a big film format if the subject is static, and if I want a lot of depth. It evens everything (focus AND resolution) out at a "poor" resolution level (not that poor, really) but the huge file, once sharpened, can print gorgeously and beat up anything in regular formats (wasn't that the same idea behind the f/64 principle?). And it's a lot less finicky when working with it.
And this also implies that you can pick up a 20$ 6x9 medium format folder with a crappy but clean triplet lens, close it down beyond f/32, shoot using a tripod, get a drum scan, sharpen the outcome, get a good print and beat 90% of present day cameras in terms of resolution on paper. It might not be practical for most people, but it can be done, and I feel it's a good thing to keep those things in mind when obsessing about resolution. The obvious downside being that physically larger sensors in the digital world are a lot more expensive.

Given that digicams have sensors with a pixel density as high as 260 lp/mm (eg the Canon G9 -- 7.6mm-wide X 4000 pixels), the 30 lp/mm values don't say much about how fine detail is rendered (260 is more than 8 times the frequency!).

Cameras like the Nikon D2x are around 90 lp/mm. The Canon 1D Mark III is around 68 lp/mm. So even for a DSLR, 30 lp/mm leaves us guessing as to what actually happens at higher frequencies. The value of 40 lp/mm (or 50 lp/mm) would be much more informative for DSLRs, but Canon only does 30 lp/mm. At least Zeiss (ZF series) does 40 lp/mm.

I display MTF cutoff for a 1.3X sensor in a similar vein in my blog entry on the Zeiss 28mm f/2 Distagon:

Puts' last article is very interesting too. The MTFs of the lenses are of relative importance. The RAW developer and software treatment can change the shape of the final image's MTF:



Dear Ruben,

Thanks for that link-- lovely data and a very nice article!

I've got two rather peripheral quibbles with the article.

The minor one is the description of practical resolution limits for handheld 35mm photography. I routinely did much, much better than that; really, by a factor of two. So did other pros I know; I'm not especially unusual that way. I'm not sure if there was a translation error on units or language, or if that's very old Zeiss data. But it doesn't jibe with my experience. Of no great import, tho'.

My second quibble is important. Erwin states at the end of his article that printers are limiting the usable resolution in the photo. Many writers have stated this; it is very wrong. An Epson 2200 printer, one of the less sharp of the modern Super-B-format printers, can usefully and visibly render at least 400 ppi detail. Most S-B printers can do better; the very sharpest ones double that. (Note: this is neither theory nor mere manufacturers' claims-- I test these things.)

A "mere" 8x10 print at 400 ppi has about 13 megapixels. To get 13 megapixels worth of real image detail out of a typical Bayer array camera would require a 30 megapixel sensor.

(Widebed printers typically have lower resolutions, but I can't imagine using my Epson 9800 to make 8x10 prints!)

I'm not asserting one needs anything like that level of detail for gorgeous, sharp prints. That would be absurd. Only that it's patently false to think that the printer is limiting print sharpness.

pax / Ctein

Dear Lloyd,

You've brought up a question that's been vexing me. What is the appropriate-size blur circle for a Bayer array camera? Is there anyone reading this who has actual, industry-experience design knowledge about this? I've read all the half-baked theories by folks who are manifestly not camera sensor/lens designers. None of them compel and all are based on highly questionable assumptions.

For square pixels of width w, I've seen people assume the smallest worthwhile blur circle to be anywhere from w up to 2.8w, with eloquent arguments made by all sides. None seem to be based on real hands-on design knowledge, though; it's lots of armchair theorizing. I am dubious.

I'd love to get an ANSWER. Or even a set of answers, if it depends somewhat upon the Bayer->RGB conversion algorithm.

Can anyone here educate me? (Please, do not direct me to YAACT*.)

pax / Ctein

(*Yet Another Arm-Chair Theorist)

Comment for Ctein:

Have you contacted Norman Koren regarding the blur circle?

He has written Imatest and Gamutvision software packages to do analysis of lenses, sensors, and printers.

With a PhD in control theory, I'm something of an armchair theorist myself, and Norman is pretty far past me in terms of photo technology.

He's been putting applying hard core math to digital cameras since about 2000 or so, and he currently writes software to do various lens and sensor analyses for companies like HP.

Check out his web-site: www.normankoren.com.


Mike, I love your site. This post is frustrating because you use an undefined initialism. What the heck does MTF stand for?

Via Bob Atkins (www.bobatkins.com), here is a link to the Canon Lens Work III:

Only people with broadband need apply!

"What the heck does MTF stand for?"

I just added a link to the first sentence of the post for you.

Mike J.

One problem with MTF charts is that they don't tell us anything about the fingerprint or signature of the lens.

As an example, I just purchased the Zeiss ZF 1.4/50 Planar for my Nikon F3-P.

Looking at the charts it promised to be an excellent performer.

In practice I am learning that while the lens is very sharp, it has a fairly ugly fingerprint.

The images it produces are very high in contrast. Lots of blacks and brilliant whites, but the middle grays are thin and feel compressed. It's very much on or off, with little inbetween.

In comparison my 2/50 Summicron-R also offers high contrast, extreme sharpness, but draws with a very evenly balanced tonal scale.

The signature of the Planar is driving me nuts to point that I went back to shooting with my old scalloped Nikkor 2/50-H.C...

Someone once told me that at Zeiss the test equipment is god and that while their lenses look stellar on paper, they may not deliver a pleasant looking image.

This must be a recent development, because the 2.8/80 Planar on my Rolleiflex delivers some of the best balanced and gorgeous images I have ever seen.

I work in the movie business and noticed the same thing, with the widely used Zeiss glass.

Zeiss Super Speed lenses are very sharp and high in contast, but deliver some of the ugliest bokeh around and a 'hard' image.

For this reason many directors of photography prefer the Taylor & Hobs primes. They are exceedingly sharp, but deliver a very smooth tonalscale and bokeh.

Try reducing the degree of film development, this will give a better grey scale. Increasing lens contrast should require a change in film contrast if you want the same gray reproduction. I don't blame a lens for having good contrast, but would blame the old lens for requiring overdevelopment to give decent final contrast.

Hey luxor2

Yes, I could change development to compensate for the increased contrast of the Zeiss. But I do change lenses on the same roll, so that's not really a solution. But you are right about the old 2/50 Nikkon HC. It is a rather low contrast optic.

But there is something else going on here. Usually I shoot with two or three bodies at a time. A mix of Leica and Nikon. The Leicas use modern glass, like the 2/50 Summicron-R and 1.4/35 Summilux-M ASPH, but also
some older designs like the 1.4/50 Summilux-M . I shoot Tri-X or Delta3200@1600 in all of them and develop in Ilford DD-X and scan on a Nikon 9000ED.

Now, the current generation 2/50 Summicron-R is a very, very sharp lens, with strong contrast, but as I mentioned earlier the tonal scale is very evenly balanced.

The images I'm getting out of the Zeiss almost have a newsprint like quality to them. I'm not an expert in optical design, but from what I understand you can emphasize contrast for certain frequencies and I believe Zeiss is doing this for course outlines, where as Leica seems
to put emphasis on increasing contrast in fine detail (micro contrast?).

In the Planar the transition from middle gray to dark or light strikes me as abrupt, above f5.6. Below f5.6 or so the lens behaves a lot better and delivers very nicely balanced images. Obviously all lenses lose some contrast when opened up, but this is pretty drastic. I've never seen such a large change in personality across the aperture range.

In comparison the finger print of the Leica Summicron is very consistent across the aperture range and it retains a good balanced across the tonal scale, while delivering a crisp, high contrast image.

It would be interesting to hear some comments on this subject from someone who has a solid understanding of optics.

The comments to this entry are closed.



Blog powered by Typepad
Member since 06/2007