Comparison of Demosaicing Methods available in Free, Open Source Raw Processors
My previous article included a table listing the various demosaicing algorithms offered by the four raw processors considered and I wondered why we (as users) needed such a wide choice. The table is reproduced below.
I decided to investigate those offered by RawTherapee by looking closely at the detail in an image of tree branches against the sky – the same part of the same raw file processed by each of the algorithms.
I’ve just gone through the same process with an Olympus Pen-F. I share the concern about JPEG file-size as a means of comparison; IME it does do a good job of compressing signal but knowing whether larger file-sizes are due to good signal or noise is difficult to establish without a visual check.
I also use entropy as a measure of information density.
It so happens that both HPHD and AmaZe are fairly middling for JPEG file-size, but they’re high on entropy. So I junked the others and retained those two.
Adding 1 step of false-colour reduction led to a massive drop in JPEG size, but again the entropies are fairly high.
When I added further detail enhancements using an optimum mixture of wavelet, light USM and microcontrast for detail, the entropy increased by 0.4% (*very* significant in this case, far outweighing the range of entropy values in demosaicing algorithms alone).
In my case, the best result was actually HPHD with extra detail enhancements as above; AmaZe came a close second. Profile saved, now I know I’ve got a good one 🙂
Given that they are quite similar in output, I am using 3.o ish version of RT because higher versions crash (I have a weird large system, 2 different video cards, 16 gigs of ram etc) And I found AMaZE to be very slow. I am experimenting with hphd 1 step now to see if it seems faster. Thanks for the analysis.
I did not specifically comment on sharpness as there is no noticeable difference in sharpness in the published (or the unpublished) images. In all cases the theoretical limit of sharpness appears to have been achieved. I agree that a larger aperture would theoretically give a sharper image but this would only be achieved in practice if imperfections of the lens did not dominate. The reduced depth of field would be a disadvantage.
I notice you make no attempt to discern any effects on sharpness. Do you not expect any?
And it might be interesting to repeat your tests on an image with a smaller airy disc.