6x lens Modulation Transfer Function curves

[A. J. deLange]

Today I took some pictures of the ISO 12233 target and analyzed the edges for chromatic aberration and MTF at the wide and telephoto extremes. I tried various edges but concentrated on the ones at the extreme corners as these should be the toughest. The chromatic aberration is excellent at less than one pixel! The MTF is very interesting in that the shape of the curve is quite different from other lenses (plot at http://www.pbase.com/agamid/image/71409544) in that the MTF starts to roll off right away, reaches the 50% point at about 350 cycles per picture height (about 700 lines pph) and then continues to decline more or less linearly remaining above 20% at 1000 lines! The reconstructed edges and the MTF's calculated from them do not show the effects of in camera sharpening which all the others do. Does the new firmware load recognize when this lens is mounted and tone down sharpening? Given these technical data I will be most interested in the group's subjective judgements of the images produced by this lens.

As always I caution people that these data represent a small subset of what a complete characterization at all combinations of focal length and aperture might show (the data sheet that comes with the lens warns of diffraction blur at small aperture) but I will chuckle silently when the usual storm about horrible CA begins.

[Chris Hurd]

Hi A.J., I split your post above out of the "6x Wide Angle Shipping" thread because I thought this was important enough that it should stand on its own. Thanks a bunch for sharing this! Much appreciated,

[Jonas Nystrom]

Sorry, but what is MTF?

My biased opinion is that 6X has far less CA the the 20X, but I haven't compared under the new firmware!

[Chris Hurd]

MTF = Modulation Transfer Function.

A parameter by which the sharpness of a lens is characterized.

See Michael Reichmann's article "The Modulation Transfer Function Explained,"
located at http://www.luminous-landscape.com/tu...ding-mtf.shtml

[Lauri Kettunen]

Originally Posted by A. J. deLange

but I will chuckle silently when the usual storm about horrible CA begins.

A.J., Thanks a lot for the contribution of posting the MTF graphs. You are such a valuable source of useful information in this forum.

What comes to this horrible CA, Digilta Film Tools (www.digitalfilmtools.com) has recently announced new software which may bit damp the noise. The 55m v7 filters for Adobe After Effects -these filters can also be imported to Premiere Pro 2.0- include two interesting tools called defringre and chromatic abberration. I tested them with my XL H1 footages and I was pretty impressed of the outcome. The undesired effects were efficiently removed.

Somebody may feel using a filter implies extra work, which is of course true, but taking into account other factors, such as the price of XL H1 etc., it's not that bad compromise compared to the final result.

[Jonas Nystrom]

Do you think that the new firmware pull down the sharpness in center of the lens to compensate for the corners? Or is it a new algoritm for the lens? Non of the others have a linear curve like as you say.

Even if the 3X which has "better start" (or?), my feeling is that the lens has an unsharpness I don't like at all (have just tried it at one job). So what is "nice sharpness" and "nice unsharpness" or blur? I think both is really important to the final picture!

[A. J. deLange]

I'm guessing that they have turned sharpening off across the whole picture as that would be easier to do than to do it for one portion of the picture as opposed to another though that would be possible as well. I've added the MTF from an area close to the center of the picture (http://www.pbase.com/image/71409544) so you can see what is probably close to the best this lens can do (near the center) as opposed to the worst case performance (at the edges). The best case performance is pretty impressive with 50% MTF (the single value most usually given) almost to 400 cycles per picture height (almost 800 lines). Perhaps the reason they turned off electronic sharpening is that this lens does not need it so much as it is capable of holding nearly 40% MTF at frequencies where the other XL lenses have responses of 10% or less.

A few comments about MTF itself: The variations in lightness and darkness of a scene modulate the amount of light which reaches a camera. The job of the lens is to transfer these modulations accurately to an electronic sensor or film. Hence the name Modulation Transfer Function which measures the relative amount of modulation transferred to the image. Engineers measure and display this data as a function of the spacing of the crests in a sinusoidal intensity modulated test image and plot the results accordingly. The reasons for doing this are pretty arcane but the result is the same as the frequency response plot for a microphone, loudspeaker, amplifier etch. In video the units of frequency are usually "cycles per picture height" which if doubled can be thought of as lines per picture height even though lines (abrupt on and off transitions from black to white) are distinct from the smooth transition from white to black and back to white of sinsusoidal intensity modulation. I mention this because the plot I have posted is arranged this way but the diagrams in the article Chris posted the link to are not.

[Jonas Nystrom]

A.J. - Really interesting post: MTF measure the lens standard in "transfer these modulations accurately to an electronic sensor or film". But the film is obsolete, when the output of chip can be manipulated a lot, so is the MTF a measurement of the lens (nowadays)? I assume it is not! For the consumer it doesn't really matter of course - and it may sound pointless - but as you say - "I'm guessing that they have turned sharpening off" - will ever know?!

It may be a stupid question, but is it at, anytime possible to get any comments from the Canon engineers about this?

[A. J. deLange]

Jonas,

Your point is a very pertinent one. No, we cannot measure the MTF of the lens alone but rather the lens in the camera where the firmware can have an appreciable effect on the system MTF.

Canon engineers do read this site but they are officially forbidden to respond.

[Lauri Kettunen]

Originally Posted by Jonas Nystrom

For the consumer it doesn't really matter

Quite so, but on the other hand the owners of the XL H1 are likely to do some of their living with the camera, and the incomplete information also blocks the users from learning more of the technology, and especially, blocks the users from learning to make the most out of the camera.

The question cannot be only of an act protecting the business, but also of something else. For instance, there's quite a lot additional information at http://www.usa.canon.com/app/html/XLH1/index.shtml which is not given in the manual. Still if you look at e.g. what is said about R-G matrix and G-R matrix, the explanation of the two is literally the same. However, the other sets the amount of (output) green in the input red channel (coming from the CCD), and the other adjusts the amount of red in the input green channel. The remaining question is, which one is which?

It's bit strange Canon fails to explain such rather trivial issues unambiguously. Especially, it's bit difficult to understand why they've incorporated all these nice things, such as the controls of colors and sharpnening in the XL H1 --notice, the owners have in fact paid for such controls-- and eventually, the users are still left in bit dark, how they could make the most out of such controls. My point is, if such things were explained, many of those consumers who do not have any technical background would still learn over time and thus gain by the information.

What comes to MTF, perhaps this site

http://www.normankoren.com/Tutorials/MTF.html

is more understandable than the one Chris mentioned.

A.J, I think I've seen a catalog containing the MTF graphs of Canon EF-series lenses. In principle, we should be able to estimate the MTF of the system by measuring the MTF of some EF-lens+XL H1 combo, and by comparing those results to the Canon given EF-series graphs. Obviously, the EF-adapter must have some effect on the MTF as well. (But, if I've got it right, there is also some adapter to Nikon lenses without any additional glass.)

[Chris Hurd]

Originally Posted by Lauri Kettunen

What comes to MTF, perhaps this site
http://www.normankoren.com/Tutorials/MTF.html
is more understandable than the one Chris mentioned.

Thanks for posting that link. Believe it or not, that's the reference that I pointed to originally in my reply above, and then I went back and changed it to the Reichmann article (linked from that site in fact) because I thought it was better written. Perhaps I should have left it on the Koren site but this way we have both references now thanks to Lauri. Much appreciated,

[Chris Hurd]

By the way, for further reading about MTF in the realm of High Definition video, check out this white paper in downloadable PDF format from Larry Thorpe, of Canon's broadcast video lens division: White Paper #3 HDTV Lens Design: Management of MTF.

This is part of an excellent and expansive series of Canon white papers covering HDTV Lens Design, the index of which can be found at http://www.usa.canon.com/html/indust...tepapers.shtml.

For example... White Paper #8 HDTV Lens Design: Management of Chromatic Aberrations.

[A. J. deLange]

Lauri,

Yes, Canon does publish MTF data for most of their lenses but not for these. It seems self defeating to me too. Do they want their consumers to have accurate data about their products as measured in their own laboratories or do they want the public's sole source of information on this sort of thing to be the results of some "kitchen chemistry" experiments by some amateur like me?

Backing out the system MTF to get a better idea as to what the lens does could be tricky, especially if they are changing internal processing when lenses are switched which they could well be doing. The best way to get good MTF data is to put the lens on a bench and use instrumentation which has an MTF appreciably better than that of the lens i.e. the MTF of the system consisting of the lens and instrumentation should be limited by the device under test. This condition would not be met in the system as we know its MTF should be near 0 at its Nuyquist frequency of 405 cycles pph.

On the subject of the color matrix: I assume that matrix entries are read with the dash meaning "to". IOW the R-G entry is the "red to green" entry which controls the amount fed from the red channel to the green channel. This could easily be confirmed by photographing something neutral and adjusting the R-G entry. If I am right increasing the R-G entry would cause a shift towards green. If I have it backwards the shift would be towards red.

[Tom Roper]

Originally Posted by Lauri Kettunen

Still if you look at e.g. what is said about R-G matrix and G-R matrix, the explanation of the two is literally the same. However, the other sets the amount of (output) green in the input red channel (coming from the CCD), and the other adjusts the amount of red in the input green channel. The remaining question is, which one is which?

It's bit strange Canon fails to explain such rather trivial issues unambiguously.

Originally Posted by A.J. deLange

On the subject of the color matrix: I assume that matrix entries are read with the dash meaning "to". IOW the R-G entry is the "red to green" entry which controls the amount fed from the red channel to the green channel. This could easily be confirmed by photographing something neutral and adjusting the R-G entry. If I am right increasing the R-G entry would cause a shift towards green. If I have it backwards the shift would be towards red.

Agree with Lauri.

A.J., then why would Canon include 6 matrix adjustments if you could do the same thing with 3? In other words, if you set G-R matrix to +3, and R-G Matrix to +3, aren't you just nulling?

[A. J. deLange]

No. If you set G-R to +3 that increases the amount of green sensor data that appears in the red channel. If you set R-G to the same value that increases the amount of red sensor signal which appears in the green channel. If you are photographing a red object, for example, increasing G-R doesn't do anything since the red object does not stimulate the green sensor. Increasing R-G takes some of the red sensor signal and puts it in the green channel so the object no longer looks red but red-green.

Actually the matrix has 9 entries including R-R, G-G and B-B which are called simply the red, green and blue gains.