Crazy LCC idea

[Shashin]

If the C1 engineers were really good, you could use the images your camera shifts to make a map of the entire image circle and then use that map to correct any image at any position--it is simply a matter of registering rasters, which could be done automatically if the cast vignetting is apparent in the image. You would have to map the image circle at different apertures, although a model could certainly be made to compensate for any aperture.

 

[GrahamWelland]

It's definitely doable. Using something like autopano pro, you can obtain a layered stitched pano, then you'd have all the masks to drop in the LCC's. That part could be automated. I know a lady wrote CornerFix for the M9, which is essentially the same thing. The processing (I mean the actual code) in cornerfix could be used to fix the layers in the stitched file.

I think Sandy might be surprised to find that he's a lady btw

I ported the CornerFix code over to a different Mac UI a few years ago and so I know this pretty well. Unfortunately it won't do what you want with LCC's for any shifted or de-centered images because of the way that it calculates the image brightness drop off and also color cast. Basically it examines the source image for the profile by calculating a color / brightness fix curve that radiates from the center of the image. This assumes a uniform color shift and vignette - not something you'll see with a shifted lens unless you determine a different offset center point. Also it won't do dust removal etc using this technique.

If the C1 engineers were really good, you could use the images your camera shifts to make a map of the entire image circle and then use that map to correct any image at any position--it is simply a matter of registering rasters, which could be done automatically if the cast vignetting is apparent in the image. You would have to map the image circle at different apertures, although a model could certainly be made to compensate for any aperture.

Indeed you could! (see above).

 

[GrahamWelland]

Rather than shooting a LCC exposure after each shot, if stitching I think shooting all the to-be-stitched images first, then shooting the associated LCCs afterwards would be preferable. The less time between stitched shots, the less change in clouds, etc... Only applicable to landscapes I guess, but that's what I mostly shoot.

Thanks for all the great suggestions everyone. VERY much appreciated.
Bob

I tend to shoot LCC's this way myself in order to minimize the shoot/adjust exposure/LCC/readjust exposure/shoot/adjust exposure/LCC/readjust exposure sequence because its easy to forget to readjust the exposure. I'll often go home with a few shots where I've forgotten to adjust the exposure back and so have to shoot the frame twice (1x idiot shot followed by the correct shot).

For stitches I'll tend to shoot the desired image sequence and then shoot the LCC's afterwards.

However, there is lot to be said for the muscle memory approach of shoot/LCC/shoot/LCC style. It's simple and easier to manage within C1 later as it becomes a mechanical process of pairing images with their adjustments.

 

[rga]

I tend to shoot LCC's this way myself in order to minimize the shoot/adjust exposure/LCC/readjust exposure/shoot/adjust exposure/LCC/readjust exposure sequence because its easy to forget to readjust the exposure. I'll often go home with a few shots where I've forgotten to adjust the exposure back and so have to shoot the frame twice (1x idiot shot followed by the correct shot).

For stitches I'll tend to shoot the desired image sequence and then shoot the LCC's afterwards.

However, there is lot to be said for the muscle memory approach of shoot/LCC/shoot/LCC style. It's simple and easier to manage within C1 later as it becomes a mechanical process of pairing images with their adjustments.

Of course there are many ways to train your muscle memory. I tend to shoot
my series (usually bracketed w/ shutterspeed) and then shoot the LCC and Color Checker last. That way I know when I look at them in C1 the final 3 LCCs belong to the preceding series.
Whatever works, as long as you can remember!!

 

[mediumcool]

Thought about this previously; you could use an LCC layered in Photoshop to achieve illumination even-ness and probably colour correction too, by judicious selection of transfer mode. Must try it one day.

 

[SCHWARZZEIT]

How do you handle LCCs on long exposures?
If your exposure is already near the limit of what the back can do, e.g. 90s for the IQ180, would you still add an additional stop or two in exposure time for the LCC?

-Dominique

 

[Guy Mancuso]

How about strobes doing interiors or something. Can't add more power. You have to open up

 

[Bob]

That is when it is appropriate to fall back on a library of standard lccs that you have previously prepared.
-bob

 

[Guy Mancuso]

Seems so

 

[EsbenHR]

If the C1 engineers were really good, you could use the images your camera shifts to make a map of the entire image circle and then use that map to correct any image at any position--it is simply a matter of registering rasters, which could be done automatically if the cast vignetting is apparent in the image. You would have to map the image circle at different apertures, although a model could certainly be made to compensate for any aperture.

Unfortunately this is not as easy as it might seem. First of all, it is surprisingly difficult to find a shift from an LCC. It looks really simple, but when you try it, the accuracy is just not very good in many realistic scenarios.

Heck, even lab conditions will drive you mad trying to get the accuracy you think *should* be in the data.

Also, the LCC does not merely records what the lens does. It records the interaction between the LCC and the sensor, which is something quite different. An LCC can only be applied to images taken with the same camera, and there is a good reason that this is the case.

 

[GrahamWelland]

Unfortunately this is not as easy as it might seem. First of all, it is surprisingly difficult to find a shift from an LCC. It looks really simple, but when you try it, the accuracy is just not very good in many realistic scenarios.

Heck, even lab conditions will drive you mad trying to get the accuracy you think *should* be in the data.

Also, the LCC does not merely records what the lens does. It records the interaction between the LCC and the sensor, which is something quite different. An LCC can only be applied to images taken with the same camera, and there is a good reason that this is the case.

This would certainly make sense. The vignette/color model used in Cornerfix suffers from inaccuracy of the correction curve estimation (taken from averaging multiple samples from radials from the selected center of the image - which for a shift would be offset - and then doing an RGB curve fit from the points) and so was never 'perfect' at the best of times, although not bad.

I hate how theory and practice never quite match up in the real world sometimes.

 

[wentbackward]

I think a library of good lcc's is not a bad approach. My biggest problem is that the light falling near the camera is often very different to the scene in front of me. And the best images always seem to be in rapidly changing conditions. I might be wrong, but i do think it actually takes quite a bit of reasonably clinical effort in controlled conditions to shoot a good profile.

Graham, I have no idea why it was in my mind that a lady engineer built corner fix. A bit embarrassing

Paul

 

[Shashin]

Also, the LCC does not merely records what the lens does. It records the interaction between the LCC and the sensor, which is something quite different. An LCC can only be applied to images taken with the same camera, and there is a good reason that this is the case.

Which is why I said you would make the LCC with your specific camera/lens combination. The software would just apply the LCC map you created. You really don't need to make an LCC for every shift position.

 

[EsbenHR]

Which is why I said you would make the LCC with your specific camera/lens combination. The software would just apply the LCC map you created. You really don't need to make an LCC for every shift position.

You can capture some information that way, but not everything. A sensor is not sufficiently uniform over its entire area to ignore these effects. The differences are normally calibrated away, but with movements you really do need to calibrate the sensor differently across the sensor.

It is, unfortunately, not generally possible to map a calibration from one part of the chip to another without sacrificing IQ.

 

[LonnaTucker]

How about strobes doing interiors or something. Can't add more power. You have to open up

Guy, opening to a wider aperture will give inaccurate LCC results, especially in the case where 2 full values more exposure are necessary.

It's easy enough in the situation you are describing to pop off a portable strobe directly at the plex covering your lens. You'd need to calculate the exposure/distance to the lens of course, but many find this produces great results. If you typically shoot at F11, you'd could repeat this without a lot of fuss.

Like Graham, I shoot the lenscast reference shots most of the time at the end of the sequence of shots for stitching. For outdoor work with light changing or clouds moving quickly across the scene, the frames for stitching need to be shot in a quick sequence for the best results and less post production headaches later on. So, having the LCC shots at the end works best for me.

In the case of long exposures, I actually reverse the process and shoot the lenscast reference shots at the beginning of the shoot, knowing that the potential buildup of noise also does not produce a good LCC file. I get the best results keeping the ISO native to the digital back and keeping the LCC exposure under 1 second. In long exposure situations, the portable strobe method works great too. (Nikon SB-900 or similar)

Lonna

 

[Guy Mancuso]

Lonna was thinking the same thing just turn the lights towards the camera and go from there. Yes i shoot the LCC after I get the image as well. Just seems to make more sense.

Great idea on the long exposures. Thanks

 

[yaya]

You need to learn how to share Yair. LOL

OK I'll share

Shot it yesterday at the Jardins des Serres d’Auteuil in Paris

Aptus-II 12R, Arca-Swiss Rm2d, SK 43mm, 10mm rise and 12.5mm shift in each direction at f8.3 and 100iso

Stitched with C1's script

 

[rga]

OK I'll share

Shot it yesterday at the Jardins des Serres d’Auteuil in Paris

Aptus-II 12R, Arca-Swiss Rm2d, SK 43mm, 10mm rise and 12.5mm shift in each direction at f8.3 and 100iso

Stitched with C1's script

Wow!
Showoff