Why Did Leica Provide Free On-Lens IR Filters?
Why Did Leica Provide Free On-Lens IR Filters?
The real question is, which of the differences (beside obviously signal/noise and resolution) cannot be leveled out or reversed in PP? willie_901, how and under which circumstances are the differences in CFA response noticeable after PP? Not trying to be snarky, I really want to know.
It can be impossible to level out or reverse caused by differences in the IR filter layer and CFA filter frequency responses.
The issue is the photon count for each R,G and B photo sites affected by differences in the IR and, or CFA frequency responses.
Consider a set of R, G, and B photosites located at the center of each sensor assembly. The final electrical charge density for each of these three photo-diodes will depend on the the sum of the filters' frequency responses. If the green CFA has significant contributions from frequencies other than green, the photon count will be in error. It will be too high. Since the correct photon count is unknown, there is no way to compute a correction. For different scenes the errors will be different because each scene will have different levels of IR, R, G and B light at these three photosites.
This is why the only complete solution to the original M8 IR contamination problem was to to supply free on-lens IR filters. The true, but unknown, amount of IR signal contamination for a given phototsite could not be computed because the spatial distribution of IR light when the shutter was open was unknowable. So, a PS Action or in-camera firmware solution was not possible.
It is true the demosaicking algorithms are designed to account for known CFA contamination effects (the overlap of R, G and B frequencies). The goal is for the demosaicking mathematical model to map onto to the data in a one-to-one fashion. When you have to correct for two different sensor assemblies the problem is complex.[1] Part of what you pay for when you license third-party raw rendering software is the R&D effort required to compute a demosaicking model that maps onto your camera's raw data.
For in-camera JPGs the issue of leveling or reversing rendering differences is moot. Besides potential in-camera demosaicking modeling differences, the JPEG compression destroys most of the original data required to level or reverse the differences.
[1] Leica deserves credit for absorbing the R&D costs required for the matching the original and replacement M9 sensor assemblies. Even so, the problem was simpler because only the IR layer filter response was different and the frequency response differences were mall and well known.
