Finally - the SD rolloff bug: FIXED!!!!!

cinefeel said:

Actually, I think the SD1 is the best sensor yet in terms of color.

Click to expand...

The thicker blue layer has to be an improvement. There were lots of problems related to the thin blue layer in the previous sensor.

cinefeel said:

And no, this gradation problem CANNOT be fixed with contrast settings.

Click to expand...

I am baffled that it seems to be able to fix with a hue setting. That seems counter intuitive to me.

--
Roland

X3F tools:
http://www.proxel.se/x3f.html
https://github.com/rolkar/x3f

Al;though I can't say that I've had this problem yet (probably due to not taking any people shots) but I just tried it, and can you confirm that your settings are in 'Color Balance' with the selector on 'Midtones' and 'Preserve Luminosity' ticked?

Otherwise the overall colour seems to be affected.

--
Thanks,
Gary.

No, you are in the wrong menu.

Don't use Color Balance.

You should be in Hue/Luminosity. This is in Image menu, after adjustments.

Make sure you drag down from 'Master' channel to the requisite color bands. The overall color balance will not change whatsoever, only the bands you specifically select.

You can fine tune the ranges if needed, but that is rarely the case I find.

HienPham said:

After all these years, they still don't understand colors.

Click to expand...

I think its a matter of philosophy.

Do you want to make correct or pleasing colors.

It seems like the earlier tries was only about pleasing colors and rendition. Bluer skies goldier yellows and avoiding clipping. And some people really do like the old look. I can even admit liking it sometimes to.

But - of course - yellow skin and DANG blue skies is nothing you want at every image. And the green plague - and the green corner plague - was rather annoying.

And one of the more surrealistic images I have seen was an Hawaii white beach that was cyan. Some claimed it looked so due to the twilight situation. Eh? Cyan twilight? No - the color was just plain out of order.

So - they decided that the pleasing colors experimenting should be over and an era of correct color was dawning. Which resulted in SD15 and now SD1. As far as I understand, those cameras are not supposed to have funky colors, but correct.

The yellow roll off syndrome shows that they failed some in that respect. They have never really succeeded in fixing a camera before. They instead come with a new tweaked camera. Lets see what happens this time.

--
Roland

X3F tools:
http://www.proxel.se/x3f.html
https://github.com/rolkar/x3f

cinefeel said:

No, you are in the wrong menu.

Don't use Color Balance.

You should be in Hue/Luminosity. This is in Image menu, after adjustments.

Make sure you drag down from 'Master' channel to the requisite color bands. The overall color balance will not change whatsoever, only the bands you specifically select.

You can fine tune the ranges if needed, but that is rarely the case I find.

Click to expand...

Ahhhhh ... I missed that ... you are tweaking bands!

Thats why it works.

Noughty boy. This is not photography - its graphics arts.

--
Roland

X3F tools:
http://www.proxel.se/x3f.html
https://github.com/rolkar/x3f

Sorry. I meant Hue/ Saturation.

And make sure you adjust Hue, not saturation.

This problem is most common in portrait shots, but also happens a lot in landscapes with leaves and foliage. The greens become polluted with this kind of unnatural yellowness, which you can remove in the same way.

If the greens are then not punchy enough, you can boost up your image as normal, having gotten rid of the yellow problem.

Is it only SD1 which is affected by this or also SD15? I have to check my photos for this. I think i saw something like this in a couple some time back.

--
Photos: http://www.flickr.com/photos/prebenr/
M42 on Foveon: http://www.flickr.com/groups/m42-foveon/

Happens in other SD cameras (actually most digital cameras) to a limited extent but I think the SD1 is the only one with a full-blown problem.

Haven't tested the older Sigmas specifically, so you might need to tweak the settings of the fix. Might work but not sure.

I think it's worse on the AFE cameras and is an issue with Sigma's 3D LUTs, which means it is a raw processing isue which can be fixed in software.

Ultimately, I am suggesting a 2D fix to a 3D problem, which is why it might occasionally need tweaking for different image conditions. Only Sigma or a third party software maker like Irrident or Ichikawa Soft can implement a truly perfect fix.

In the meantime, I'm happy with this and have made it into an automated photoshop action!

cinefeel said:

Sorry. I meant Hue/ Saturation.

And make sure you adjust Hue, not saturation.

Click to expand...

So - essentially what you do is you move the red and yellow bands to another color.

Ouch!

Did you read my post earlier regarding correct colors vs pleasing colors philosophy?

Tweaking like this is according to the pleasing colors philosophy. You do whatever makes your boat float. Thats nice for tweaking, but works rather badly for making a good general solution. Sure - you avoid yellow skin - but you might make red cars pink.

--
Roland

X3F tools:
http://www.proxel.se/x3f.html
https://github.com/rolkar/x3f

No, I've tried it in images with red in them- even a red car in fact- and there is no problem.

You are making the assumption that the colours are more correct in the first place, and that this fix is moving them to an 'incorrect' place.

I would say the exact opposite is happening!

My image of the dog shows this. This fix makes the colors MORE correct, not less. it just shows that Sigma made a mistake with their raw processing LUTs.

You can also see on my image of the face that the blotchy noise 'magically' disappears! Your can see the same kind of color noise on various SD1 images, regardless of the actual image colors, and this fix makes it disappear

Actually, I would say that the SD1 in neutral mode and recent firmware is an extremely color accurate camera. Infinitely more so than older Sigmas.

cinefeel said:

No, I've tried it in images with red in them- even a red car in fact- and there is no problem.

Click to expand...

OK - good

cinefeel said:

You are making the assumption that the colours are more correct in the first place, and that this fix is moving them to an 'incorrect' place.

Click to expand...

Nope - I make no assumptions whatsoever. You make assumptions regarding me making assumptions

cinefeel said:

I would say the exact opposite is happening!

My image of the dog shows this. This fix makes the colors MORE correct, not less. it just shows that Sigma made a mistake with their raw processing LUTs.

Click to expand...

Might be so - you might be lucky.

cinefeel said:

You can also see on my image of the face that the blotchy noise 'magically' disappears! Your can see the same kind of color noise on various SD1 images, regardless of the actual image colors, and this fix makes it disappear

Actually, I would say that the SD1 in neutral mode and recent firmware is an extremely color accurate camera. Infinitely more so than older Sigmas.

Click to expand...

Still you have to apply your fix

OK - lets see if I can explain my view. Its the correct vs pleasing colors philosophy.

PLEASING COLORS

In the pleasing colors philosophy you do tweaking of the color matrix in order to make something you do appreciate. Now ... +7 on the orange HUE slider - and now it looks nice! This is a in principle a dangerous method. Sure - you might be pleased with what you see. But ... it might not work for other images and you might be pleased by something else tomorrow.

In your tweaking you took yellow and red and moved those color bands. I dont know in what direction - but lets assume you moved yellow towards orange. Then maybe two different input values both might result in the same orange hue. Thats called metamerism. And you can do other distortions of the color space.

I think this is how Foveon/Sigma was working up before SD15. Thats why their iamages tended to be green, magenta and cyan. Strange color casts that normally other cameras dont do.

CORRECT COLORS

Correct colors is a completely other strategy. Then you have to make correct color management. You have to make transforms that for a chosen illumination produces "correct" colors.

NOTE - this is hard for Foveon sensors as there is no correct transform - only approximations. So - the approximation have to have some measure of goodness.

This is the right way to approach the problem IMHO. Then you shall use some kind of calibration hard and software.

--
Roland

X3F tools:
http://www.proxel.se/x3f.html
https://github.com/rolkar/x3f

Roland Karlsson said:

PLEASING COLORS

In the pleasing colors philosophy you do tweaking of the color matrix in order to make something you do appreciate. Now ... +7 on the orange HUE slider - and now it looks nice! This is a in principle a dangerous method. Sure - you might be pleased with what you see. But ... it might not work for other images and you might be pleased by something else tomorrow.

In your tweaking you took yellow and red and moved those color bands. I dont know in what direction - but lets assume you moved yellow towards orange. Then maybe two different input values both might result in the same orange hue. Thats called metamerism. And you can do other distortions of the color space.

I think this is how Foveon/Sigma was working up before SD15. Thats why their iamages tended to be green, magenta and cyan. Strange color casts that normally other cameras dont do.

CORRECT COLORS

Correct colors is a completely other strategy. Then you have to make correct color management. You have to make transforms that for a chosen illumination produces "correct" colors.

NOTE - this is hard for Foveon sensors as there is no correct transform - only approximations. So - the approximation have to have some measure of goodness.

This is the right way to approach the problem IMHO. Then you shall use some kind of calibration hard and software.

Click to expand...

I doubt there exists any camera digital or film that can take correct colours( ). Post is allways about making a more "pleasing image/photo". But as to an automated approach to fix this problem I agree with you that the proposed method cannot be auto-applied to all photos. However I think the problem arises in the processing of the RAW data. As with the purple highlight in SD15 firmware. So there might be a firmware tweak to be applied.

( ) correct as in how our eyes percieve colour frequencies...
--
Photos: http://www.flickr.com/photos/prebenr/
M42 on Foveon: http://www.flickr.com/groups/m42-foveon/

Sorry about the bold, should have written (#)

Annoying that one can only edit a post once in the archaic forum software of dpreview

--
Photos: http://www.flickr.com/photos/prebenr/
M42 on Foveon: http://www.flickr.com/groups/m42-foveon/

Well it may help in some (or in a lot) of case but it is no fix. Although it often may result in a pleasing photo it introduces quite some other problems. For example if a photo has real yellows in it like a yellow jacket or similar, it turns it into some sort of orange. Overall I've found the photos end up being more "red", a bit more monotone. From what I see you basically bring the reds and yellows closer together, shifting it to more or less one color. This results will of course minimize the roll-off phenomena but will add problems elsewhere.

I wish there was a way for us to play with DNG profiles to see if maybe using more complex profiling could solve this "problem".
--
Bob van Ooik
V-studio
----------------------------------------------
http://vstudio-magazine.de/
http://www.x3magazine.com/
http://www.flickr.com/photos/bobnl/

By the way, I have been using this fix on images with the SPP color wheel set to zero.

I'm not sure if it's different based on the color wheel settings.

Certainly nothing is perfect in the world of digital cameras and color, especially something as relatively untested as Foveon. All I'm suggesting is that this tweaking of the hues creates a better starting point for post processing, compared to the SPP defaults.

The range of the color bands are tweakable and I have not tested extensively what the best is. You can try it yourself and see what suits your images. If desired, you could even apply the fix selectively on problematic parts of your image!

cinefeel said:

By the way, I have been using this fix on images with the SPP color wheel set to zero.

I'm not sure if it's different based on the color wheel settings.

Certainly nothing is perfect in the world of digital cameras and color, especially something as relatively untested as Foveon. All I'm suggesting is that this tweaking of the hues creates a better starting point for post processing, compared to the SPP defaults.

The range of the color bands are tweakable and I have not tested extensively what the best is. You can try it yourself and see what suits your images. If desired, you could even apply the fix selectively on problematic parts of your image!

Click to expand...

I remeber some portraits of my youngest daughter and I got very ugly yellow skin tones. I'll test with similar aproach in Aftershot/Bibble5 (have to figureout if it is possible) to see how it plays out. Sometimes even after setting the whitebalance there are problems with the colours being shifted or somewhat off... But I have not tried this before.

--
Photos: http://www.flickr.com/photos/prebenr/
M42 on Foveon: http://www.flickr.com/groups/m42-foveon/

BobNL said:

For example if a photo has real yellows in it like a yellow jacket or similar, it turns it into some sort of orange.

Click to expand...

perfect.........

BobNL said:

I wish there was a way for us to play with DNG profiles to see if maybe using more complex profiling could solve this "problem".

Click to expand...

agree...........

BobNL said:

--
Bob van Ooik
V-studio
----------------------------------------------
http://vstudio-magazine.de/
http://www.x3magazine.com/
http://www.flickr.com/photos/bobnl/

Click to expand...

--

To understand photography, you must understand that the experience must be much more important than the result ....
Carlos Roncatti Bomfim
http://weweh.com/
because men and women are different and we humans are the same.

PrebenR said:

I doubt there exists any camera digital or film that can take correct colours(*).

Click to expand...

You dont have to doubt, its a fact

But - different cameras do a different good job. Bayer cameras has a good start by detecting color bands that are quite near the correct ones. Foveon has some totally different starting point and needs much more transformation.

PrebenR said:

Post is allways about making a more "pleasing image/photo". But as to an automated approach to fix this problem I agree with you that the proposed method cannot be auto-applied to all photos. However I think the problem arises in the processing of the RAW data. As with the purple highlight in SD15 firmware. So there might be a firmware tweak to be applied.

Click to expand...

Agree

--
Roland

X3F tools:
http://www.proxel.se/x3f.html
https://github.com/rolkar/x3f

BobNL said:

Well it may help in some (or in a lot) of case but it is no fix. Although it often may result in a pleasing photo it introduces quite some other problems. For example if a photo has real yellows in it like a yellow jacket or similar, it turns it into some sort of orange. Overall I've found the photos end up being more "red", a bit more monotone. From what I see you basically bring the reds and yellows closer together, shifting it to more or less one color. This results will of course minimize the roll-off phenomena but will add problems elsewhere.

Click to expand...

This is not a surprise

BobNL said:

I wish there was a way for us to play with DNG profiles to see if maybe using more complex profiling could solve this "problem".

Click to expand...

Some did calibration of Sd14 or SD15, probably SD14.

As far as I remember they mad calibrations with regard to a known illuminant.

A more useful calibration would be to make a model of white balance and make one calibration of the camera. Dont know if its possible for an X3 color space though.

--
Roland

X3F tools:
http://www.proxel.se/x3f.html
https://github.com/rolkar/x3f

To BobNL:

You can avoid the yellow/red = orange problem if you select the band ranges carefully in Photoshop. You can also use Yellow -10 Red -+0 as an alternative, or as desired for your particular images.

I would also restate that my particular fix settings currently assume that your white balance is correct and you have left the SPP color wheel at zero.

Either way, you can simply use this fix selectively in the rolloff areas only, by using layers in Photoshop. This is probably the best idea until Sigma creates a deeper fix.

However, I would say that they cameras rendering of yellow at the moment is currently incorrect. I have seen this in my own photographs of yellow things.

Also for example, Bob, the yellow jacket and skin tones in your recent Hip Hop dance duo shot looked unnaturally yellow to me. Interestingly, the embedded jpegs looked normal- suggesting that the mapping problem is in SPP and not in the camera.