philip_pj wrote:
"..like all demosaicing algorithms includes a sharpening process."
Ken, not to question the veracity of this statement, but I saw this statement recently, again in c't:
"Sharpening is most often performed after demosaicing...The Raw Therapee documentation confirms this assumption and Sony confirmed it when asked. However, Canon, Nikon and Adobe refused to give us any information on the subject."
Hi Philip,
All I meant, but didn't say specifically, is that any demosaicing algorithm includes something akin to sharpening in its basic demosaicing algorithm. Some software will do a separate step of what we'd traditionally consider "sharpening" as well that you may or may not be able to turn off or adjust. Even without this additional step the various demosaicing algorithms affect high spatial frequencies differently, they have a "native" amount of sharpening within the algorithm's kernels even if it isn't immediately obvious as a separate sharpening step.
Demosaicing is at its heart a form of interpolation, all interpolation involves filtering kernels and sharpening is just a high pass filtering kernel. If different demosaicing kernels have different high pass characteristics this is very much like a difference in sharpening between the two kernels. Convolve a sharpening or bluring kernel with your "native" demosaicing kernel from one algorithm and it will look a lot like the kernel from a different demosaicing algorithm.
Many of the algorithms with weaker "native" sharpening look suprisingly comparable to algorithms with higher "native" sharpening once sharpened to match. This is not to say all algorithms are the same by any stretch, rather that sometimes the differences are masked by simple sharpening effects that are easily equalized across different demosaicing algorithms. This is clear if you do a lot of comparisons between ACR/LR and other converters. You can achieve a wide range of sharpening effects in ACR and even vary the sharpening algorithm with the detail slider (100% is pure deconvolution, 0% is almost pure USM). Once you've tweaked ACR sharpening much of the detail differences with other RCs disappear and you are left with more interesting residual differences that are a bit more subtle.
As others pointed out detail and sharpness is only one part of what makes a RC "good" but sharpening has a very strong affect on perceived detail. Get that out of the way and you can focus on color processing, noise character, highlight and shadow recovery, etc. etc. etc.
You are definitely right about "the more you know the less you know", I feel that way the more I look at and try to understand practical image processing! In general every simple assumption I try to make eventually proves to be insufficient.
Ken