Good show JACS, though the 200% images are one step removed from what my comments have been about throughout this thread: 4 capture pixel shift vs single capture, which remove effective pixel pitch as a variable and thus can be indicative of the difference in performance of Bayer CFA vs monochrome sensors. The difference can be hard to spot in the conditions shown earlier.
On the other hand, doubling the number of linear captures, pixel pitch and hence image size as in your example just above does indeed increase rendered resolution as well as reduce aliasing further, as mentioned upthread. The effect is quite obvious.
One more note on B&W honesty before leaving this interesting subject.
J A C S Wrote :
Patterns on the cloths and on the faces look good, actually, better than the demosaiced versions I have seen. So, this would be the honest representation of a B&W image if we knew it is B&W.
It is 'a' B&W representation. And this begs the question of what would be an honest one.
One way to answer it is to assume that honest B&W is what a monochromat would see, let's say in daylight conditions. Let's further assume that the monochromat only has the cone type that is sensitive to Medium wavelengths. That means that such a person effectively sees something proportional to Luminance, because that cone type has spectral sensitivity practically equivalent to the relative luminosity function (V).
On the other hand a typical Bayer CFA sensor has three types of color filters with peaks in what we normally refer as the R, G and B wavelengths - none of which are even close to the luminosity function. But let's say that we take the G channel as being equivalent to V, thus honestly (?) taking care of about half of the pixels in our B&W rendering. After all, the G plane is usually where cameras set their focus on.
The other half of pixels, R&B, have sensitivity functions that are even further removed from V. But let's give them the benefit of the doubt: after white balancing, if the illuminant is uniform and the scene is neutral, we could get intensity somewhat proportional to the others, right?
The R and B pixels are however on different color planes. That means that they are most likely out of focus and more aberrated (e.g. lateral chromatic) compared to G. So what we get is truly a Frankenstein of a B&W rendering, which its proud parents try to pass on as the real thing. Honest
Jack
PS By these standards a pixel shifted render is not very honest either. In fact no render with commercial sensors is. But of course they are good enough for most purposes and we are all happy with the results.
