Fuji Super CCD non-linearities

[Guillermo Luijk]

From this thread about the Fuji Super CCD, I have done a test developing the RAF file provided with strict camera white balance in the whole dynamic range, and it seems confirmed the reason why some Fuji users have reported magenta cast in the very high lights with some RAF files: the R sensor in the Fuij Super CCD seems to have some non-linear behaviour in the last f-stop of the DR.

The RAW file had no levels blown, I used a correct saturation point. I applied a strict camera white balance in the whole DR with DCRAW (-H 1). Other programs such as ACR/LR force a neutral WB in the highlights, that's why probably the user's render does not show this issue because I get the same result with DCRAW's -H 2.

While most of the image is right, the highlights (see window's floor and border) were not properly balanced appearing magenta cast:






There is a clear misalignment between the RGB channels in the last f-stops which according to the scene should be close to neutral (R=G=B ). In particular there is a lack of G values:




The conclusion could be that it is recommended not to take Super CCD to its limits or problems may arise when using certain RAW developers (such as Fuji HU). Other sensors are more linear in this last f-stop before saturation.

[David Ellsworth]

Hi GLuijk,

I extracted the R-pixel image with "dcraw.exe -4 -T -D -j -t 0 -s 1 DSCF0028.RAF". This extracts the image rectangle tilted 45 degrees, with no interpolation.
Then, in Photoshop, I extracted the green channel from it as follows:

• Edit->Preferences->General: Image Interpolation: set to "Nearest Neighbor"
• select Marquee tool
• Select All
• Free Transform: 45 degrees horizontal skew
• Image->Image Size: resample to 50% horizontal size (Nearest Neighbor)
• Select All
• Free Transform: -45 degrees (negative 45 degrees) vertical skew
• Edit->Preferences->General: Image Interpolation: set to whatever it was set to before
• If you want to get rid of the excess black, then select it with the Magic Wand Tool at Tolerance=0 without anti-aliasing, Invert Selection, and Image->Crop.

(Incidentally, the above procedure can be used to extract the green channel from Bayer RAWs; of course you leave out the "-s 1" option in dcraw. The green channel comes out tilted 45 degrees, but uninterpolated. To prevent cropping, you need to enlarge the canvas before Step 1.)

Here is the extracted R-pixel green channel, at full size and not gamma-corrected.

Now, notice that the brightest highlights in the green channel have a disproportionate amount of noise! (I see the same effect in this photo's R-pixel red and blue channels.) This is because the camera's amplifier/ADC doesn't clip R-pixels at saturation; it allows them to reach full well capacity, where evidently strange nonlinearities occur. This is like what Sigma Foveon cameras do in their RAW files. (Although the Fuji also does a nasty thing: it doesn't add a bias level, thus clipping the shadows a tiny bit.)

As for how to solve this... that'd require some study. I tried clipping the R-pixels to a lower saturation point; unfortunately this also clips some areas that have real detail. It seems that the R-pixels start to lose some charge, randomly, after being heavily saturated. It'd help to have a set of "saturation frames" in which the camera is aimed at something bright and uniform, and successively overexposed.

Well, just to show something, here's what I get when I clip the R-pixel image enough to get rid of all this strange highlight noise:
dcraw.exe -T -s 1 -S 5000 -r 1.8 1 1.4 1 -H 1 DSCF0028.RAF

[Guillermo Luijk]

here's what I get when I clip the R-pixel image enough to get rid of all this strange highlight noise:
dcraw.exe -T -s 1 -S 5000 -r 1.8 1 1.4 1 -H 1 DSCF0028.RAF

yes, definitively there is no useful information in the magenta areas of the R-fuji, so it's best to clip them to pure white.