Lenses for D800e

[marcmccalmont]

I would look at DxO Mark rankings
Marc

[erpman]

Of the nikon AFs the 1.4G appears to be the best one overall. But as you say, it´s not on their recommended list. Ken Rockwell claims that the 55mm 2.8 AF micro is the sharpest one produced by nikon.

[bjanes]

Of the nikon AFs the 1.4G appears to be the best one overall. But as you say, it´s not on their recommended list. Ken Rockwell claims that the 55mm 2.8 AF micro is the sharpest one produced by nikon.

I am perplexed as to how Nikon compiled that list. The AFS f/4 24-120 VR is on the list, but it has not received good reviews from photozone.de, Lloyd Chambers (pay site), and others. It performs well in the center of the field in the lower focal length range, but the corners are soft and it has marked distortion and light fall-off. Digilloyd says it performance is very poor at 120 mm and does not consider it a professional grade lens, although it is priced as one. It's too bad, since it would otherwise be a good walk around lens. I have ordered the 800E, but I will stay away from the 24-120 VR. I already have the AFS 60mm f/2.8, which is on the list and I have found to be a very sharp lens with good contrast, both for closeups and distance. If you don't need auto-focus, the Zeiss Makro Planar 100 f/2 is very highly recommended; I have considered buying it, but it is rather pricey and I already have the 105 MicroNikkor 2.8D.

Regards,

Bill

[JeffKohn]

So - at F11 you already have a size of the first diameter of the diffraction disc which is big enough to swallow 4 pixels.
This means every single light ray is blurred to 4 pixels!
This alone will drop the effective resolution of a 36 MP image of the D800 to something below 9 Megapixels.

Your theoretical calculations vastly overstate the real-world effect of diffraction. I'm not saying that diffraction is a non-issue, but to say that it reduces high-res cameras to an effective 9mp at f/11 is just silly.

[Christoph C. Feldhaim]

Your theoretical calculations vastly overstate the real-world effect of diffraction. I'm not saying that diffraction is a non-issue, but to say that it reduces high-res cameras to an effective 9mp at f/11 is just silly.

I'd be happy if you could show me where the flaw in my thinking is.
Since my thoughts are deductive in nature from given facts there must be a flaw if you find them silly.

[jeremypayne]

You may find this interesting ... scroll to the bottom:

http://www.luminous-landscape.com/tutorials/resolution.shtml

[Christoph C. Feldhaim]

You may find this interesting ... scroll to the bottom:

http://www.luminous-landscape.com/tutorials/resolution.shtml

You have all the data at hand, but take the green-yellow light and f/8-f/11 aperture values as a reference. It represents a realistic, not too demanding case. Consider a 35mm system with a lens at f/11. At best, the maximum resolution you will get is equivalent to 16 MP, even if your camera has 22 or 25 MP.

Seems I wasn't too far off with my 9 Megapixels, though it was a bit too pessimistic.

[billy]

Of the nikon AFs the 1.4G appears to be the best one overall. But as you say, it´s not on their recommended list. Ken Rockwell claims that the 55mm 2.8 AF micro is the sharpest one produced by nikon.

thanks. are you referring to this lens? ( http://www.bhphotovideo.com/c/product/585343-USA/Nikon_2180_AF_S_Nikkor_50mm_f_1_4G.html )  it is not in stock at BH Photo.

I would not consider the 55mm 2.8AF micro for my type of work, it is too slow.

[erpman]

That´s the one. The lens-comparison function at DxoMark is really comprehensive and useful. Take all the 50mms and run them up against each other, see which one suits your needs. The 2.8 is not fast, thats true, but it is sharp. I borrowed one once for macro work and it was a joy to use.

[erpman]

I'd be happy if you could show me where the flaw in my thinking is.
Since my thoughts are deductive in nature from given facts there must be a flaw if you find them silly.

The flaw in this particular context is that you´re not taking into account that stitching 20 frames from a tele-lens increases resolution dramatically, thereby reducing the degree of interpolation/reduction in print resolution necessary to blow up the image. Theoretically you can print 2x3m at 300dpi without interpolating when stitching (as long as you have a good computer). Therefore, the diffraction at a given pixel, when seen in context of the huge size/resolution and viewing distance of the final print, is negligible.

[Christoph C. Feldhaim]

The flaw in this particular context is that you´re not taking into account that stitching 20 frames from a tele-lens increases resolution dramatically, thereby reducing the degree of interpolation/reduction in print resolution necessary to blow up the image. Theoretically you can print 2x3m at 300dpi without interpolating when stitching (as long as you have a good computer). Therefore, the diffraction at a given pixel, when seen in context of the huge size/resolution and viewing distance of the final print, is negligible.

Sure.
I simply fail to understand why one would need a 36 MP camera when using F11 or F22 most of the time.
Will record a lot of data junk ...

[Bart_van_der_Wolf]

I simply fail to understand why one would need a 36 MP camera when using F11 or F22 most of the time.
Will record a lot of data junk ...

Hi Christoph,

While I agree it (a high samping density sensor array and lots of diffraction blur) is not the most logical combination, it still works reasonably well at lower ISO settings in combination with deconvolution sharpening. The diffraction pattern produces pixels with a weighted average response with weighted contributions from surrounding pixels. That oversampled diffraction pattern can be relatively (compared to defocus) well restored to the central pixel's contribution alone.

Cheers,
Bart

[Christoph C. Feldhaim]

Hi Christoph,

While I agree it (a high samping density sensor array and lots of diffraction blur) is not the most logical combination, it still works reasonably well at lower ISO settings in combination with deconvolution sharpening. The diffraction pattern produces pixels with a weighted average response with weighted contributions from surrounding pixels. That oversampled diffraction pattern can be relatively (compared to defocus) well restored to the central pixel's contribution alone.

Cheers,
Bart

Hi Bart,
How much would you estimate could be restored then? I even tend to shoot not above f8 to preserve detail in the area of focus with my Mama 7. I don't care too much about out of focus areas, since for me the blur from oof areas is part of every photograph. But sometimes I would like to use the higher f stops too.

Cheers
Chris

[erpman]

Sure.
I simply fail to understand why one would need a 36 MP camera when using F11 or F22 most of the time.
Will record a lot of data junk ...

You just fail to see how diffraction relates to print resolution in this context. Sure, f.22 looks horrible on a single frame on any sensor but thats not what we´re discussing. When you quadruple the resolution and sharpen properly, the loss of detail due to diffraction is vastly overtrumped any day. With 36m you can also stitch, but with fewer images and a wider lens=more dof=larger aperture=less diffraction. Print size=the same.

[Bart_van_der_Wolf]

How much would you estimate could be restored then?

Hi Chris,

It depends on the image quality (noise, and optical residual aberrations).

Ideally one would create a proper model of the lens' point spread function (PSF) at the given apertures, for a given sensor design. A good PSF approximation can be created by shooting a test chart (e.g. like this) which also has slanted edges. Those slanted edges allow to synthesize a pretty good PSF. One then needs to use software that can use such a custom PSF as input (I use ImagesPlus for problem images).

Lacking that, we can do a less accurate attempt by using an iterative approximation with a program like RawTherapee which offers 'Richardson-Lucy' deconvolution as a sharpening method. Other possibilities are deconvolution sharpening plugins like FocusMagic (which unfortunately is lacking support for 64-bit applications), or TopazLabs' "Infocus" (which needs to be improved a bit to reduce artifacts).

If you post a crop of an image at 2 different apertures, say f/5.6 and f/16 (or f/22), I'm sure there will be a few readers who want to give it a try and see what can be restored. If you want to have a better result, then shooting a target is required, from which I could make a PSF to be used.

Cheers,
Bart

[bwana]

it seems the zeiss 25mmm f2 has some of the highest mtf in the charts.
http://www.lensrentals.com/rent/nikon/lenses/wide-angle/zeiss-zf.2-25mm-f2-for-nikon
hwever, the 21mm is often cited as legendary (must be for other qualities)
http://www.lensrentals.com/rent/nikon/lenses/wide-angle/zeiss-zf.2-21mm-f2.8-for-nikon

would these not be better than the 14-24 zoom nikkor? and certainly a lot less stitching than using the 100 mm(though some distortion correction in pp)

[bwana]

A pixel of the D800 has a pitch of 4.88µ.
The diagonal of a square of 4 pixels is (2*4.88)*sqrt(2)=13.4µ.
13.4µ corresponds to an f-stop of 13.4/1.35=9.9 at green light.
So - at F11 you already have a size of the first diameter of the diffraction disc which is big enough to swallow 4 pixels.
This means every single light ray is blurred to 4 pixels!
This alone will drop the effective resolution of a 36 MP image of the D800 to something below 9 Megapixels.

if you consider a light ray as a single bit of information, then a 4 pixel patch on the sensor is this information bit.
but consider that each pixel is the sum of the corners 4 adjacent patches-the information bits overlap.
and each information bit is the result of a different light 'ray'- so the overlapping information bits are sightly different.

i think we can calculate the unique and accurate value of individual pixels by solving a set set of simultaneous equations-
consider each pixel on the sensor as a unique variable. 4 pixels(a,b,c,d for example) are grouped into an equation (a+b+c+d=light value)
there are as many equations as there are pixels.
therefore there is a unique solution using matrix algebra that solves for the values of the individual pixels.

at least that's how i would solve the diffraction problem. is this solution flawed? has it been done?
solving 36 million equations in 36 million variables should not be too hard-computer graphics cards and cpus operate in gigahertz frequencies (billions and billions as Carl Sagan said about stars)

PS: sorry for a reply to an off topic fork that started in this thread but a lot of people seem to be participating in it. if you feel a different thread is needed, let me know and i will do so.

[Christoph C. Feldhaim]

You just fail to see how diffraction relates to print resolution in this context. Sure, f.22 looks horrible on a single frame on any sensor but thats not what we´re discussing. When you quadruple the resolution and sharpen properly, the loss of detail due to diffraction is vastly overtrumped any day. With 36m you can also stitch, but with fewer images and a wider lens=more dof=larger aperture=less diffraction. Print size=the same.

Well, larger aperture (F-Stop below F11) surely counters my argument, but that was not what I was going to tell.
So we have:
Case 1: High resolution (D800), wider lens and lower F-Stops, or
Case 2: Lower resolution Camera (D3 or so), narrower tele lens, higher F-Stop and more  stitches.

Cheers
~Chris

[Christoph C. Feldhaim]

Hi Chris,

It depends on the image quality (noise, and optical residual aberrations).

Ideally one would create a proper model of the lens' point spread function (PSF) at the given apertures, for a given sensor design. A good PSF approximation can be created by shooting a test chart (e.g. like this) which also has slanted edges. Those slanted edges allow to synthesize a pretty good PSF. One then needs to use software that can use such a custom PSF as input (I use ImagesPlus for problem images).

Lacking that, we can do a less accurate attempt by using an iterative approximation with a program like RawTherapee which offers 'Richardson-Lucy' deconvolution as a sharpening method. Other possibilities are deconvolution sharpening plugins like FocusMagic (which unfortunately is lacking support for 64-bit applications), or TopazLabs' "Infocus" (which needs to be improved a bit to reduce artifacts).

If you post a crop of an image at 2 different apertures, say f/5.6 and f/16 (or f/22), I'm sure there will be a few readers who want to give it a try and see what can be restored. If you want to have a better result, then shooting a target is required, from which I could make a PSF to be used.

Cheers,
Bart

Hey thanks !
I always like reading your explanations.
Target printed.
Rest will follow

Cheers
~Chris

if you consider a light ray as a single bit of information, then a 4 pixel patch on the sensor is this information bit.
but consider that each pixel is the sum of the corners 4 adjacent patches-the information bits overlap.
and each information bit is the result of a different light 'ray'- so the overlapping information bits are sightly different.

i think we can calculate the unique and accurate value of individual pixels by solving a set set of simultaneous equations-
consider each pixel on the sensor as a unique variable. 4 pixels(a,b,c,d for example) are grouped into an equation (a+b+c+d=light value)
there are as many equations as there are pixels.
therefore there is a unique solution using matrix algebra that solves for the values of the individual pixels.

at least that's how i would solve the diffraction problem. is this solution flawed? has it been done?
solving 36 million equations in 36 million variables should not be too hard-computer graphics cards and cpus operate in gigahertz frequencies (billions and billions as Carl Sagan said about stars)

PS: sorry for a reply to an off topic fork that started in this thread but a lot of people seem to be participating in it. if you feel a different thread is needed, let me know and i will do so.

Its not a linear problem I believe due to the nature of the PSFs, thats why matrices would be limited here if I guess right.
If I'm  not mistaken the reversal problem is done differently by the deconvolution algorithms  - I believe Bart could explain it much better than I ever could how it works mathematically.

Cheers
~Chris

[erpman]

Hi Chris,

It depends on the image quality (noise, and optical residual aberrations).

Ideally one would create a proper model of the lens' point spread function (PSF) at the given apertures, for a given sensor design. A good PSF approximation can be created by shooting a test chart (e.g. like this) which also has slanted edges. Those slanted edges allow to synthesize a pretty good PSF. One then needs to use software that can use such a custom PSF as input (I use ImagesPlus for problem images).

Lacking that, we can do a less accurate attempt by using an iterative approximation with a program like RawTherapee which offers 'Richardson-Lucy' deconvolution as a sharpening method. Other possibilities are deconvolution sharpening plugins like FocusMagic (which unfortunately is lacking support for 64-bit applications), or TopazLabs' "Infocus" (which needs to be improved a bit to reduce artifacts).

If you post a crop of an image at 2 different apertures, say f/5.6 and f/16 (or f/22), I'm sure there will be a few readers who want to give it a try and see what can be restored. If you want to have a better result, then shooting a target is required, from which I could make a PSF to be used.

Cheers,
Bart

I took the liberty of playing around with one of my own. This is shot at canon 5dmkII with canon 70-200 f4 L @160mm @f16. The red square marks the area of the two 100% crops. They are processed in RawDeveloper with just a slight WB and contrast adjustment. One has no sharpening, the other has been sharpened with R/L Deconvolution at radius .60 and 25 iterations. The third one has gotten additional sharpening in Photoshop.

The one without sharpening is definitely soft, and some softness is of course because of the AA filter in 5dmkII. I am however fairly happy with the quality of the sharpened one (RLD and RLD/ps), especially when constitutes a couple of centimeters in an image that prints about 1x1,5m at 300dpi without interpolation in PS.