I have a Fujifilm FinePix S200EXR which uses a “Super CCD EXR” sensor. In contrast to the Bayer square layout, it uses a diagonal sensor pixel layout where some sensor pixels can be “binned” to optimize for either a larger dynamic range or for a better signal-to-noise ratio in low light scenes.
Due to this rather unique layout, I wonder if any special demosaic algorithms would be recommendable. Or are the diagonals just hiding that there is still a Bayer pattern, and I don’t need to worry?
As a new member, I am not sure whether I am able to link images yet… but you will surely find several examples searching for images by keywords like “super ccd exr sensor” or similar.
RT supports extracting frame one or two (both having 6 MP) and the SN-Mode where the two frames are combined to reduce noise (not for more resolution). DR mode is not supported by RT.
I am quite disappointed to learn that different EXR cameras seem to produce different raw image content, related to their operating mode. My S200EXR creates RAF HD with two images, and only Fujifilm and Adobe seem to know how to (re-?) construct 4000×3000 pixels from their resolution of 2824×2128 stored raw pixels. Applying SN averaging where resolution interpolation was desired is not an optimal solution.
In contrast to other reports, RAF SN and DR created by the S200EXR have only half the file size of HD; dcraw reports only one image, so I have to assume they contain an already averaged image in SN mode, or an already combined HDR result in DR mode.
I linked a sample archive in your HDRMerge github project.
I tried with this raw which is from an S200EXR in whatever mode (it contains two equaly exposed 6 MP frames). RT definitely can access both 6MP frames in this file and also can combine them.
Your image has a size of ~24 MB. So it must have been shot in HR mode (not in SN mode).
If I take a shot in SN mode, the RAF file has only a size of ~12 MB. So I have to assume that the camera already did the averaging and saved one raw image with the averaging result.
If I take a shot in HR mode, the RAF file has a size of ~24 MB. So I have to assume that the image is not supposed to be averaged, but to be interpolated to a high resolution instead.
I wish I knew how to interpret RAF header data to see if any flag in it might give a hint in which mode the camera was set up.
From RT perspective, it doesn’t matter what’s in the exif of EXR files concerning shooting mode.
If there are two frames in the raw, RT can access each of them and also average them. Not more, not less. If the two frames have different exposure (for example this file) averaging them does not make sense, but you still can do it, because there are two frames in the raw.
Set up a custom conversion preset: DNG 1.4 + Linear (mosaic free) — and voila, RawTherapee 5.6 reports: 24.0 MP (5640x4248) … from here, the camera apparently scales down to 4000x3000.
Probably not. It may have a Super-CCD sensor, but not in the HR mode of the EXR models. I see no choice between EXR operation modes in the S5 Pro manual. To see whether S and R pixels are stored separately or combined as one pixel with higher precision, and if the diamond shape results in two frames, one may need a raw file to compare meta values and structure. I don’t have any.
Well, interesting: This raw image contains 2 sub-frames. Extracting them with dcraw -T -j -s all returns two images which differ slightly in their exposure and color balance, so I would assume they are S and R sub images. But what’s really special: They are stored diagonally (tilted by 45 degrees), so they are obviously stored according to the diamond-shaped pixel layout of the SuperCCD sensor. As displayed in the English Wikipedia about SuperCCD.
In contrast, the sub images of an S200EXR are stored straight aligned, but with the half-pixel diagonal interleave shift of the binned pixel pairs.
Linear (mosaic free) — and voila, RawTherapee 5.6 reports: 24.0 MP (5640x4248) … from here, the camera apparently scales down to 4000x3000.