Here's what the M8 firmware does for red vignetting

[scott kirkpatrick]

Skip this thread if you don't have patience with graphs and numbers, but there is information in Sean Reid's white wall pictures that gives a pretty good picture of what the M8 firmware does to correct color vignetting, and what the sources of this are for the 28mm lenses that he discusses in his current draft article at reidreviews.com.

Here are two plots, for the Elmarit 28mm and for the Summicron 28mm, of the relative strength of the red compared to the average of the green and blue values along a line starting at the middle of a uniform white rectangle and proceeding to the upper right corner. In each case the lower lines are without any firmware correction, and the upper pair of lines are with lens detection enabled (these are Leica-coded lenses).



Notice the results for f/2.8 and for f/8 are pretty much the same. The Elmarit is a more symmetric design than the Summicron, and sits closer to the CCD surface. So it shows stronger vignetting. The firmware increases the strength of the reds near the edge of the frame to the same value (depressed by about 8%) by making a stronger correction for the Elmarit than for the Summicron. This firmware was developed before Leica recommended that IR filters be used, so it is the correction for the extra blue-green filtering that occurs at the edges due to the existing cover glass on the CCD. The white walls were shot with an IR cut filter added to the front of the lens. In principle, the red vignetting expected from a filter in front depends only on the angle of view (of the focal length) of the lens, and should be the same for the symmetric Elmarit and the telecentric Summicron. These graphs are consistent with that expectation.

After looking at the roughly equal size of the color vignetting corrections from the filter and from the cover glass, I would want to have the assistance of the firmware both with and without an IR filter. I think there will need to be a menu item toggling the correction for an IR cut filter off and on. Just turning lens detection on and off is too crude to give the best color fidelity.

scott


PS -- a few details. The measurements were made by taking screen captures of the four unlabelled white wall shots with and without lens detection in the current version of the article. I then folded my laptop screen flat, facing up and placed a marked piece of paper across each image diagonal to determine the locations to measure. (The last points are unreliable, because I couldn't measure a constant distance from the corner.)

I also have measurements of overall vignetting for these two lenses, which do depend on aperture as well as the details of the lens design. I'll discuss them after comparing with the uncoded lenses. There is no accurate theory for this, since it is measured after including the helping effect of the offset microlenses.

[ho_co]

Scott--
Extremely interesting! Shows exactly the degree of planning Leica had done in recognizing the problems wide-angles would impose! I'm amazed at the quality of thinking they demonstrate; their products work because of extreme technical know-how that most of us would be completely unaware of without people like you making this kind of effort to uncover it.

Very good job greatly appreciated!

--HC

[sean_reid]

"Scott...most of us would be completely unaware of without people like you making this kind of effort to uncover it."

Hmmm.... ???

It will be interesting to look at what differences we'll see in the new firmware and also at the effects of the in-camera processing on hand-coded lenses (as per Scott's, I mean my <G>, upcoming tests).

Cheers,

Sean

[sdai]

I really appreciate your effort, Scott ... but with due respect, the usefulness of these on the fly tests is very doubtful because the error in measurement is totally out of your control. There's no way I'll believe the light falloff is zero even at approximately 25% away from the center of the 28 cron ... just for the record, here's the factory vignetting chart of both lenses. Stopping down from f2 to f5.6 does improve a lot for the cron ... curiously, f2.8 down to f8 doesn't make much difference for the 28 elmarit asph ... so how much help the 6-bit coding could lend in this particular case. Sorry ... I just don't get it.

[marknorton]

Scott, certainly very interesting. I especially like the way you are plotting R relative to B+G to normalise for the overall sensor vignetting correction.

If I understand you correctly, these shots are with IR filters and show that the M8 already has some correction for red vignetting which is effective in that the corrected results for both lenses are the same. The residual fall off is then caused by the IR filter which is largely independent of both lens type and aperture.

It would be interesting to know how close to flat the results are without an IR filter. I agree with you the lens recognition menu will need to specify whether or not an IR filter is used and so allow additional correction. Still an open question whether different makes of filter cause different levels of vignetting.

I'm wondering whether there is at least some correction applied even if lens recognition is turned off?

The whole lens coding idea is at last being shown to be delivering value - no thanks to Leica who are altogether too vague about detailed technicalities.

Interesting piece of work, thank you.

[marknorton]

Simon, the key point is that 2 * R/(B+G).

It's plotting red relative to blue/green and therefore "filters" out the vignetting due to the lens alone and the overall correction applied due to sensor vignetting. This additional fall-off is due to IR filtering alone.

IMHO, it's a good piece of work.