Upload & Sell: Off
| p.33 #9 · Still no love for the Ricoh GXR? |
Kosmos; Have you tried any of the shorter summarits? They're supposedly "better" with digital sensors than the older elmarits of similar focal lengths.
Yes, RPP - and a few others - do make a lot of difference. Mind that a raw image is (in this case) some 12 million pixels, and that for each pixel you need to estimate two "supposedly correct" values for the other two channels.
A very fast (but still accurate) edge sensing 3x3 algorithm will need about 20 ops and 4 memops (I don't count register ops, they're so fast) to calculate every new color value, giving you about 35-40 cycles per pixel - times two channels, times 12 MP >> a bare minimum of ~100 million calculations per raw image.
Now go JUST ONE step beyond this, and check a 5x5 grid around the pixel you're currently evaluating, and the amount of calculations increase by about a factor of ten.
The 3x3 grid isn't enough "statistics" for each pixel to get rid of colour aliasing effects, and not nearly enough to get smooth lines (unless you have an AA filter, then it works with accuracy perfection), but a 5x5 - or even larger - will at least enable the converter to get rid of COLOUR aliasing. You can't touch the luma aliasing, unless you go to AI pattern recognition schemes, and then we're talking about hours per image.
uscmatt's images does probably contain a lot more "contrast per pixel width" across the thin lines in the house facade, considering the light conditions. But those images would also benefit from a better raw converter... I don't like aperture, for many reasons. The two things it has going for it are a slick user interface and actually quite brilliant highlight clipping handling.