Nikon D800 is here...

[fvdbergh]

I am actually curious about how Bayer filter demosaicing would work if you remove the LP filter --- this would imply that you cannot interpolate between the neighbouring photosites, because they are not correlated.

Thom Hogan mentioned Rob Galbraith's article:

I've been having a hard time tracking down an absolute answer on what that D800E filter does. Rob Galbraith wrote that it blurs verically and then deblurs, while the non-E model blurs horizontallly and then blurs vertically. I've gotten conflicting answers out of Nikon sources, but I believe Galbraith is correct: the front stage of the filter on the D800E is part of an optical system in the filter itself.

Rob's article (http://www.robgalbraith.com/bins/multi_page.asp?cid=7-11674-12304) includes some diagrams, but it is not entirely clear whether neighbouring photosites (of the same colour) will still be somewhat correlated, or not.

I have seen MTF50 values greater than 0.25 cycles per pixel on unsharpened images, which implies that normal LP filtering gives us resolution between 0.5 cycles per pixel (upper limit from sampling theory) and 0.25 cycles per pixel (lower limit if LP filter ensures neighbouring photosites of same colour are very strongly correlated). The peak MTF50 values I have seen are about 0.28-0.3 cycles per pixel, and Norman Koren mentions values of 0.33 cycles per pixel.

The D800E could in theory squeeze into that space, i.e., give us resolution between 0.3 cycles per pixel and 0.5 cycles per pixel. It would be very interesting to see exactly where it ends up in this range.

 

[Dolby]

What is the link between the normal resolution cameras and the high resolution - and the current lenses?

I see a few of you have mentioned new lens or different quality?
The more megapixels, the better the lens needed?

 

[fvdbergh]

What is the link between the normal resolution cameras and the high resolution - and the current lenses?

I see a few of you have mentioned new lens or different quality?
The more megapixels, the better the lens needed?

The sensor cannot resolve more detail than what the lens can capture. There may be some benefit to increasing resolution beyond what the lens can capture, e.g., Bayer demosaicing artifacts (colour Moire) can be reduced, but smaller pixels have reduced capacity for capturing photons, so photon shot noise will increase. So my take is that you want to keep the pixels as large as possible, but small enough to resolve all the detail captured by the lens.

Comparing, for example, the peak centre sharpness of the Nikkor 35 mm DX f/1.8 G to that of the Nikkor 24-70 mm f/2.8 (the lens used in some of the sample D800 shots on Nikon's site), using LensTip's measurements, tells us that the 24-70 mm lens is slightly sharper (about 8%, so larger than experimental error). This means that the 35 mm f/1.8 lens could not max out the resolution of a 10 MP sensor, which may be a bit troubling, since this would suggest that the 35 mm lens would not be able to exploit the additional resolution of the D800.

I have another interesting comparison: how does my copy of the 35 mm f/1.8 perform on a D7000 relative to the D200 used by LensTip? I have measured about 55 line pairs per mm peak sharpness. It is a little hard to compare this to the LensTip results because we use different software, but with a metric ton of salt, we can go ahead and compare the results anyway. The difference in linear resolution between the D200 and the D7000 is about 25%, which means that extrapolating the LensTip D200 results on the 35 mm f/1.8 should give use 57 line pairs per mm, which is quite close (but beyond the expected experimental error) to what I have measured on the D7000. The D7000 pixels are almost the same size as the D800, so we will probably see 55 lp/mm resolution on the D800 with this lens.

So on the one hand, we see that really expensive lenses (e.g., 24-70 mm f/2.8) produce higher resolution than a decent cheap prime (35 mm f/1.8), but on the other hand we see that the cheap lens continues to improve with higher resolution sensors (but we expect performance to level off at some point).

My guess is that the pro Nikkor lenses will be good enough for the D800. There may be a little bit more resolution that could be squeezed out of the D800E, but we will have to wait for reviews. Or someone can donate a D800E for testing ... ... but don't forget to include some pro lenses!

 

[Quantum Theory]

Or someone can donate a D800E for testing ... ... but don't forget to include some pro lenses!

Ja, send some this way too... I don't mind shooting some sample landscapes for free.

 

[fvdbergh]

I have another interesting comparison: how does my copy of the 35 mm f/1.8 perform on a D7000 relative to the D200 used by LensTip? I have measured about 55 line pairs per mm peak sharpness. It is a little hard to compare this to the LensTip results because we use different software, but with a metric ton of salt, we can go ahead and compare the results anyway.

I have made some more careful MTF measurements on the Nikkor 35 mm f/1.8G @ f/4. This time, I used a soot-blackened razor blade as measurement target (not as convenient as a printed target, but the black is more uniform than toner-on-paper, and the edge is arguably straighter). In the centre of the lens, I measured 56 line pairs per mm, which is close to the 57 lp/mm expected on the D7000, which we get from extrapolating the LensTip results on this type of lens on a D200.

Conclusion: a cheap prime lens can be sharp enough so that it is limited by sensor resolution on e.g. a D200. Sharpness measurements of the same lens on a higher resolution sensor, e.g. a D7000, are higher by almost the same factor as the linear increase in resolution (D200 -> D7000 ~ 25% increase). We can expect comparable resolution out of the D800 using this lens, i.e., the lens is "good enough" for the D800.