Add new comment

Losing all benefits of Sigma (Foveon X3F) Merrill BW benefits

One of Foveon's sensors used in five different camera models:

  • SD1
  • SD1 Merrill
  • DP1 Merrill
  • DP2 Merrill
  • DP3 Merrill

doesn't permit the inherent BW quality advantages normally provided for these cameras when using LibRaw. As an example, BW images only open as colour images in LibRaw.

This issue was escalated to the LibRaw dev team by the Affinity (Serif) devs about a year ago, but I couldn't find it here, so I am raising the issue again in the hope of a solution.

The above sensor has three layers of overlapping sensels for the same three Cartesian coordinates (no CFA pattern is used). In other words, for any given X,Y sensor location, you can obtain all RGB approximations without needing to apply any form of demosaicing. Within this physical silicon sensor, the Blue layer is on top, followed by Green, and the Red layer is lowest: this means that the Red layer is prone to the greatest noise, similarly to someone diving deeper into the ocean, and finding that the Red part of the spectrum has the hardest time penetrating the medium.

For many years, this layered sensor topography has delivered a great advantage for BW photography: you can add the layers to obtain a relatively panchromatic spectrum for every single sensel point (again, no demosaicing required). Since these sensors also feature no antialias/low-pass/blur filter, they not only can skip that demosaic step, but they can also forego all sharpening, and because three separate sensor captures are used (no pixel shifting or multi-exposures), you get benefits of sharp exposure averaging, which means more accuracy and smoother output. When factoring all these advantages together, the output is just like one obtains when using a Monokrom or a specially modified sensor (one where the Bayer and Blur filters have been removed), and in fact, usually better.

There's a few ways to do this when opening a BW Sigma/Foveon RAW image (X3F format):

  1. Treat the image as panchromatic, and add together the counts for each sensel's three RGB layers, for a linear brightness, almost like CMOS binning, and also providing triple exposure averaging.
  2. Do similarly to the above scenario, but permit independent adjustment of each RGB contributing layer (i.e., when simulating BW filters: to obtain an amber filter effect, include greater portions of Red and Green, but less Blue layer).You get more artistic control, at the expense of some exposure averaging.
  3. The least desirable method, at the expense of forcing the spectral response of always using a Blue filter, only uses the Blue layer sensel counts: you miss out on the benefit of exposure averaging from the other two layers, but at least you are using the top-most (least noisy) layer.

I am not able to do any of these three methods using LibRaw: something happens to them during import that throws away that Foveon goodness, making the images look harsh. I have been pursuing this for well over a year with Affinity, but they assure me that the issue is entirely with LibRaw, and since several camera bodies share this same sensor and experience the exact same problem, I am hoping a common fix can be made to let these models behave more like other Sigma/Foveon sensors. For example, I don't have this LibRaw issue with other Foveon sensors (such as the one used in the SD14 camera).

Image icon AffinityPhoto, using LibRaw247.33 KB