Jump to content

CornerFix for Windows available for download


sandymc

Recommended Posts

Advertisement (gone after registration)

Dear folks, it is real great analysis here. And Sandy is suggestion is for sure most welcome. But pardon on my ignorance, what do you mean by off-centre lens? Is it a lens manufacturing defeats or something else.

 

First, we have the focus shift and now we have the off-centering on the world reowned 35mm lens, must be kidding....

 

Well, not even Leica is perfect(!) All that I'm referring with the off center discussion to is that you would expect that the point of near-zero light fall-off (the maximum on Scott's curves) would be in the center of the sensor - in the vertical dimension, at +- at row 1300 on the sensor. Actually, in one case I've seen the maximum at about 1100. Name of lens manufacturer not provided, so as to protect the guilty.....

 

This can be inherent in the lens itself, or in the lens mount, or in misalignment of the sensor within the body. Mostly, actually, this makes very little difference - certainly on a film camera, you wouldn't notice, just because luminance vignetting is actually quite difficult to see in practical photographic situations. But color casts are easy to see, which is why its an issue on the M8.

 

Sandy

Link to post
Share on other sites

  • Replies 96
  • Created
  • Last Reply

Top Posters In This Topic

Top Posters In This Topic

Posted Images

Just a query, I was planning to make a reference image for my CV21mm as a starter. How critical is the actual colour of grey card required and is outdoor daylight aceptable to illuminate, etc. What size grey cards are available in the UK? I have only seen Kodak 18% card advertised but that is A4 which does not seem large enough for a 21mm where I reckon I would need more like A2 for the reference image to get a full frame coverage. My local art shop sells card in all colours which I use for mounting photos, Dove Grey being among them, etc.

George

Link to post
Share on other sites

George,

I asked the same question and it isn't important it just needs to be evenly lit and fairly neutral. Download the version from yesterday and check out the updated FAQ section in the quick start PDF. Sandy's updated it with answers to all the questions I had as well as some others.

 

- Carl

Link to post
Share on other sites

Just a query, I was planning to make a reference image for my CV21mm as a starter. How critical is the actual colour of grey card required and is outdoor daylight aceptable to illuminate, etc. What size grey cards are available in the UK? I have only seen Kodak 18% card advertised but that is A4 which does not seem large enough for a 21mm where I reckon I would need more like A2 for the reference image to get a full frame coverage. My local art shop sells card in all colours which I use for mounting photos, Dove Grey being among them, etc.

George

 

George,

 

The actual color isn't critical at all, as long as its fairly neutral - in fact, a white card is just as good - I've used just a sheet of blank paper......

 

Likewise, the light isn't critical - as long as the resulting image has a histogram that shows a reasonable amount of light in all channels, no blown highlights, you're fine.

 

Sandy

Link to post
Share on other sites

on c), I don't know if centering is so easy to detect in amateur lab work like I have tried. The illumination differences are hard to control, and I would prefer to only apply corrections which have been averaged over the four corners, as CornerFix presently does. But having two sliders, one for strength of the overall correction, and the second for strength of the red vignetting correction, where the test data choose the functional form, is a great idea.

 

And I'm happy to keep the images coming. No ACK required, as you might be inundated right now.

 

scott

 

Scott, yes, it's quite possible that the current averaged over all four corners correction will remain the best choice. Just some preliminary analysis is showing that while the centering issues is real, its not going to be easy to deal with. What's clear is that vignetting is coming from different sources, and each of those sources is centered differently. In the M8, there are at least four sources of vignetting:

 

1. Classic optical vignetting

2. Vignetting from the sensor

3. Vignetting from the microlenses

4. Mechanical vignetting

 

Below are two graphs, one showing vignetting not corrected for centering, and the other corrected. This is for a zoom lens from a major Japanese manufacturer, btw, not an RF mount lens. What's interesting is how at the extremes, the graph kinks, and becomes straight, probably as a result of mechanical rather than optical vignetting. And that you can center for the first part of the graph (optical vignetting), but that centering on that just makes the kinked part (mechanical vignetting(?)) worse.....

 

The various lines are vertical, horizontal and diagonal cut across the sensor, btw.

 

Sandy

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

Link to post
Share on other sites

 

What's clear is that vignetting is coming from different sources, and each of those sources is centered differently. In the M8, there are at least four sources of vignetting:

 

1. Classic optical vignetting

2. Vignetting from the sensor

3. Vignetting from the microlenses

4. Mechanical vignetting

 

 

Let's see if I am on the same wavelength, by checking terminology. The classic definition of vignetting (disregarding our unique color-effects) talks about angle dependence being cos^3 theta due to the shrinking area of the aperture seen at an angle and the greater distance from aperture to sensor -- that would be #1? Then there is another factor of cos theta in film due to the decreased absorption of the light by sensitive grains when it enters the film at an angle. Something like that also decreases the sensitivity of silicon photosensors -- #2? The microlenses are offset by an amount that is calculated to be tan theta for some most interesting focal length, which we don't know in this sensor -- has anyone read what it is? #3? I would think that 1-3 form a common source of vignetting, and can't be separated or are not worth separating. And finally, most long fast lenses have extra vignetting when used at or near widest aperture, because the corners of the frame can't see the whole aperture -- #4?

 

That leaves #5, which would be angle-dependent absorption of the long-wavelength parts of the red signal. Incidentally, I'd like to see how different IR-rich light like tungsten or household light bulbs behaves when analyzed with your tools.It's much harder to get uniform lighting on a target with this sort of stuff, but the relative fraction of long-wavelength light is much more, so the correction curves should be different.

 

scott

Link to post
Share on other sites

Advertisement (gone after registration)

Let's see if I am on the same wavelength, by checking terminology. The classic definition of vignetting (disregarding our unique color-effects) talks about angle dependence being cos^3 theta due to the shrinking area of the aperture seen at an angle and the greater distance from aperture to sensor -- that would be #1? Then there is another factor of cos theta in film due to the decreased absorption of the light by sensitive grains when it enters the film at an angle. Something like that also decreases the sensitivity of silicon photosensors -- #2? The microlenses are offset by an amount that is calculated to be tan theta for some most interesting focal length, which we don't know in this sensor -- has anyone read what it is? #3? I would think that 1-3 form a common source of vignetting, and can't be separated or are not worth separating. And finally, most long fast lenses have extra vignetting when used at or near widest aperture, because the corners of the frame can't see the whole aperture -- #4?

 

That leaves #5, which would be angle-dependent absorption of the long-wavelength parts of the red signal. Incidentally, I'd like to see how different IR-rich light like tungsten or household light bulbs behaves when analyzed with your tools.It's much harder to get uniform lighting on a target with this sort of stuff, but the relative fraction of long-wavelength light is much more, so the correction curves should be different.

 

scott

 

Scott,

 

Yes, (1) is the classic cos(a1)^3cos(a2), where a1 and a2 are the lens "entrance" and "exit" angles respectively - cos^4 for a theoretical thin lens.

 

(2) is about the CCD elements being effectively in "pits" - what literature I've managed to track down indicates that the vignetting should be approximately linear with exit angle, but the work relates to CCD sensors from a long time ago.

 

And (3) is effectively unknown. Some literature indicates that the microlenses effectively offset by an angle of the order of about 17-20 degrees in DSLRs, but what effect they have on vignetting in an M8.......well, I guess Leica know, but I sure don't

 

BTW, in multiple equation mode, CornerFix does try the whole gamut of cos^4, cos^3*cos, etc, etc equations, but in practice, none of the cos type equations have to date been a good fit onto any real lens.....

 

Agree it would be very interesting to compare say flash-lit vignetting with tungsten-lit.

 

Sandy

Link to post
Share on other sites

George,

 

The actual color isn't critical at all, as long as its fairly neutral - in fact, a white card is just as good - I've used just a sheet of blank paper......

 

Likewise, the light isn't critical - as long as the resulting image has a histogram that shows a reasonable amount of light in all channels, no blown highlights, you're fine.

 

Sandy

Hi Sandy, I've just made my first profile. Assuming that I have done it OK - I get odd concentric circles on my modified DNG with the CV21 + 486 filter. Used 'antique white' and diffussed sunlight to photograph a dng to make profile. Would you like me to mail you the profile and before & after DNGs. Assuming of course that I have not done a silly - in which case - what?

Hope you can help me

George

Link to post
Share on other sites

Hi Sandy, I've just made my first profile. Assuming that I have done it OK - I get odd concentric circles on my modified DNG with the CV21 + 486 filter. Used 'antique white' and diffussed sunlight to photograph a dng to make profile. Would you like me to mail you the profile and before & after DNGs. Assuming of course that I have not done a silly - in which case - what?

Hope you can help me

George

 

George, yes, send the image and profile along to me. I'll PM you an e-mail address.

 

Sandy

Link to post
Share on other sites

George, yes, send the image and profile along to me. I'll PM you an e-mail address.

 

Sandy

 

I have sent the information 'off group', thanks. I shall be interested in the results, probably me being a thicko.

George

Link to post
Share on other sites

I have sent the information 'off group', thanks. I shall be interested in the results, probably me being a thicko.

George

 

No, not you being a thicko - in fact a genuine problem, that someone else has also come across in an image strikingly similar to yours - lots of blue sky. There's a solution under test right now.............if anyone else sees this problem, feel free to PM me.

 

Sandy

Link to post
Share on other sites

c) It may be worthwhile to introduce two new abilities into CornerFix - firstly, the ability to manually adjust for centering (a suggest from Carsten), and also it may be possible to add the capability to scale back on corrections, so not 100% correct, similarly to what Leica do. Perhaps that could be done separately in the luminance and color channels, so you could 100% correct for red vignetting, but not 100% for luminance vignetting.

 

 

Sandy, just coming back to this (after all this tech talk): From a normal photographer's point of view, your program works perfectly. Corrections are almost too good for real life pictures with a wide angle lens, where one is used to some degree of vignetting in the corners and first thinks that CornerFix is over-correcting, even it is not. For this reason, I created profiles at F8, but the additional possibility to scale back on the correction for luminance vignetting would be great.

 

Priority at this time, however, should really be on batch processing.

 

Wolfgang

Link to post
Share on other sites

Here are some results from the Elmarit 24/2.8 shooting in daylight. The first figure below traces out the diagonals across the frame when shooting with a B&W486 filter at f/2.8, comparing uncorrected RGB values, corrected using CornerFix and this image to generate a profile, and less strongly corrected using a shot taken at f/5.6 to generate a profile. The vignetting and the relative drop in the red signal is a bit stronger for this lens than what I posted earlier for the 35/2.0. The full correction profile appears to over-correct slightly in the bottom of the image (the right side on my plots), but the weaker profile removes almost all vignetting, without overdoing it. By contrast, the "ON+UV/IR" correction, shown in the next graph, neutralizes the red/green differences, but reduces the overall vignetting only slightly. I attach the original shot and the corrected shot make with the weaker profile, below the curves.

 

scott

 

edit: the second image posted below is the +UV/IR corrected frame, which shows considerable vignetting. I will attach the better corrected shot done with CF and the milder profile in the next post

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

Link to post
Share on other sites

Here's the white wall shot, corrected in CornerFix with the profile generated at f/5.6. That profile is available here, and the profile generated at f/2.8 is here. Both will be available with other CF profiles whenever a good host is located.

 

scott

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

Link to post
Share on other sites

Here are two images taken the other day in my local supermarket with a CV21 at about f8, to test the lens since it had just been delivered.

The first is processed with ACR4.1 and the second with Cornerfix initially ( a second version Sandy sent me to evaluate). I only get feint concentric circles of a sky picture, again as a test shot.

Some difference in the corners

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

Link to post
Share on other sites

Here are the concentric circles I refer to, rest assured that I do more than photograph the sky. It was taken to judge the vignetting on my new CV21 with B+W486. Sometime of course some photos will include a lot of sky as part of its composition. I have sent this direct to Sandy btw. There are a couple of other things about usability I have also sent direct to Sandy, ie need confirmation which profile is in use and flicking between file with profile & the image file. Btw it is really easy to make a profile for those to try, I used a sheet of A2 Antique White card propped on my garden seat in bright diffused sunlight with camera on a tripod. That was my first tryout. I'll do my CV25 and my old trusty 35mm Summicron (which is not on the Leica list for 6 bit coding btw) another day.

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

Link to post
Share on other sites

In response the the concentric circles problem above, beta 0.9.0.4 is now available in the usual place:

 

SourceForge.net: CornerFix

 

This version reduces the problem that George is referring to above relative to 0.9.0.3, but does not entirely eliminate it. Unfortunately, the problem is as a result of the 8-bit compression scheme that M8 DNGs employ. The way that current releases of CornerFix work is to decompress the DNG, correct for vignetting, and then recompress to 8-bit. At the moment, it does not appear that the concentric circles problem can be completely eliminated while still maintaining the 8-bit recompression process. So a good chance that to completely eliminate this issue, future releases of CornerFix will have to output full 16-bit DNGs, rather than 8-bit DNGs. :mad:

 

So consider this is an interim partial fix...

 

Sandy

Link to post
Share on other sites

I am puzzled by this posterization effect. I don't see it in the examples that I have posted above, and they are pure highlight, so should be vulnerable to posterizing. The strongest corrections gave a slight shoulder to each graph, that was not seen in the slightly weaker corrections. Is that an artifact?

 

Does mixing the three colors overcome it, and only when you get a pure color like blue you see steps? BTW, it is very hard to see this effect in the posted web-size pictures that are just above here. How about introducing some dithering?

 

scott

Link to post
Share on other sites

I am puzzled by this posterization effect. I don't see it in the examples that I have posted above, and they are pure highlight, so should be vulnerable to posterizing. The strongest corrections gave a slight shoulder to each graph, that was not seen in the slightly weaker corrections. Is that an artifact?

 

Does mixing the three colors overcome it, and only when you get a pure color like blue you see steps? BTW, it is very hard to see this effect in the posted web-size pictures that are just above here. How about introducing some dithering?

 

scott

 

Scott, it is indeed an artifact, and you'd only see under quite rare circumstances - the only actual examples that have emerged so far are scarily similar - both on CV21's, and both blue sky. The reason why blue sky is where this shows is the relative sensitivity of the sensor; the red channel is a lot less sensitive than blue (or green), and Leica don't correct for that prior to compression. It pops up badly when red is about 1/2 to 1/3 the value of the blue channel......just about exactly sky color(!)

 

0.9.0.4 doesn't dither, but does use some anti-aliasing code that in effect gives the same result.

 

Sandy

Link to post
Share on other sites

I am puzzled by this posterization effect. I don't see it in the examples that I have posted above, and they are pure highlight, so should be vulnerable to posterizing. The strongest corrections gave a slight shoulder to each graph, that was not seen in the slightly weaker corrections. Is that an artifact?

 

Does mixing the three colors overcome it, and only when you get a pure color like blue you see steps? BTW, it is very hard to see this effect in the posted web-size pictures that are just above here. How about introducing some dithering?

 

scott

 

The circles you see here are much reduced from the previous version, you can see them if you are looking for them. As a photograph it is done for beta testing of a product. I doubt they would be seen in say street photography, I'll process some others I took on the same day when I get a chance and see if the 'circles' are visible. If anyone would like a copy of these as DNGs or PSD files to see the effects full size please PM me with an email addy and I can send them on.

George

Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...