Navigation

May 9, 2012

LumoLabs: Nikon D800 AA filter vs. D800E

MTF curves for the Nikon D800 and D800E SLR full frame cameras.
Taken in the center with a Nikkor 24-70/2.8G lens at 50mm f/4.0 and critically controlled perfect focus, using studio flashes. Images taken RAW and processed in Lightroom PV2010 with default settings except for no noise removal and different sharpening settings: D800 sharpened with 1.0 px radius and 100% amount, D800E sharpened at 0.8 px radius and 70% amount. This results in both images having the same 10-90% edge rise width (1.32 pixels) and almost the same MTF50 resolution.


Ever since Nikon announced that the D800 will come in two flavors, D800 with a Bayer-AA filter and the D800E without one, people wondered how large the difference would be and what version to get.

Obviously, the D800 is meant for APS-C and 35mm full frame SLR crossgraders as all current such cameras with a Bayer sensor have a Bayer-AA filter. While the D800E is meant for medium format (or Leica M9) crossgraders as all current such cameras have no Bayer-AA filter. The latter makes sense because of the high resolution of the D800/E which at 36 MP touches what once was medium format territory (40+ MP).

However, don't believe for a second that Nikon created the two versions for technical reasons. They have not. They created two versions to serve two separate markets. It is this simple. So, don't expect Nikon has any clue what camera would be the better one for you. All they'll try is ask you questions to figure out which market you belong to. Not their business, not really helping in your decision.
If you want to know what's really in the boxes and how it really differs, you have to measure it. 

Which is exactly what my friend Dieter Lukas from panobilder.de and myself did and now want to share with you. You obviously need both a D800 and D800E to study the differences ;)

Results

We have prepared a detailed report about our findings and testing procedures which you are invited to read:
In a nutshell:

We determined the exact MTF of the Bayer-AA filter in the D800 and deduced the strength of the beam splitter (which such an AA filter really is). We determined its strength in the Nikon D800 to be rather weak, around 75% of what would be a full strength filter.

As a consequence, the difference between a D800 and D800E isn't as large as one may think: in a controlled environment, the D800 images can be sharpened to the level of the D800E. The downside is that it can produce some false colors too, although less likely and to a lesser extent.

As a rule of thumb, we found that (assuming 100% amount, in Lightroom terms) subtracting about 0.5 px from the sharpening radius used for a D800 image produces comparable sharpness and acceptable results. In practice, one may of course combine this with a larger radius and lower amount etc. We summarize this into the following headline:
D800: E = 0.5px sharper

Meaning, that with ~100% amount sharpening, the D800E should deliver comparable results with ~0.5 pixels less sharpening radius, compared to a D800. This also means that one should not refrain from sharpening when using the D800E. Just use weaker settings.

Examples

Below are two test chart 100% crops for the D800 and D800E, using the different sharpening settings as described above. To compare samples with identical parameters, please refer to the full report.

Nikon D800 sharpened with 1.0 px radius and 100% amount in LR3.
The edge blur width is 1.33 px.

Nikon D800E sharpened with 0.8 px radius and 70% amount in LR3.
The edge blur width is 1.30 px. The Nyquist limit is near the figure denoted "6".
As you can see, the results are pretty similiar, with a bit more sarurated false colors and false color moiré in the D800E (as to be expected). But the D800 is able to show a bit of false color moiré too (a phenomenon known from the Canon 5DmkIII too).

In the real world, we found false color moiré from the D800E not to be problem. We only spotted it in a very few shots so far and here is one rare example where it occured to us in the wild:
100% crop from a D800E image exhibiting a false color moiré pattern.
(Please, click on the image for the full size original image)

.
Your mileage may vary if you shoot man-made patterns (fabrics, buildings) for a life. The choice is up to you but it is our impression that either camera represents an excellent choice where differences don't matter as much as some may think (this statement includes video applications too).


Thanks for stopping by,

Dieter & Falk

31 comments:

  1. This comment has been removed by a blog administrator.

    ReplyDelete
    Replies
    1. Removed because Anonymous w/o nickname, in a foreign language, off-topic and written in an aggressive and destructive style. This is not a forum, please behave adequately, Thanks.

      Delete
  2. Ja, das mag stimmen, hat aber üerhaupt nichts mit dem AF-Problem zutun....

    Es geht nur darum in wie Weit sich die beiden Versionen der D800 unterscheiden und zwar im Bezug auf ihre Schrärfe....

    Hat sich da gerade ein Nikon-Benutzer auf den Schlips getreten gefühlt?

    Im gesamten Beitrag steht doch NIRGENDWO "AF" oder "Autofocus"

    Den Beissreflex muss man sich erstmal angewöhnen.

    Thx @ Falk and Dieter für die hervorragende Arbeit und das ganze geteste.

    ReplyDelete
  3. You can also sharpen with D800 E to extend where D800 can't reach...0.5xD800E > 0.5D800

    ReplyDelete
    Replies
    1. Show me the RAW files.
      Having fone extensive testing in RAW and found no such thing.

      Delete
    2. and your name is?

      And if you would read before you write, you would have seen that your comment was premature.

      Delete
  4. How has your choice of testing a zoom lens vs. a high quality fix focal length lens somewhat muted the differences between the two cameras?

    ReplyDelete
    Replies
    1. That's a good question, but no.
      The 24-70/2.8G at 24mm actually performs on par with the best prime lenses, esp. in the center and at f/4. At 50mm, it is a tad weaker but not to the extent that I expect the differences between cameras to become muted. Also unlike the photozone test, our 24-70/2.8G at 50mm shows best center resolution around f/3.5-f/4, a behaviour which normally only good primes do show.

      So, although being a zoom, the center at 50/4 clearly outresolves the sensor.

      Delete
  5. Salute to your hard work.
    what you are saying is to sharpen 100% in 1 px will do the work?
    same in photoshop?

    ReplyDelete
    Replies
    1. For an otherwise sharp D800 photo, yes. Photoshop has several sharpeners and their scales do differ. In many situations, there is additional image blur and stronger sharpening settings may provide better results. OTOH, in the presence of image noise, less sharpening may be more appropriate. YMMV.

      Delete
  6. First thank you for the hard work and for sharing the results. Please consider making the original NEF files available.

    ReplyDelete
    Replies
    1. For a limited time, I'll grant access to raws at
      http://www.falklumo.com/lumolabs/articles/D800AA/res/DSC_{0746|0192}.NEF

      Enjoy :)

      Delete
    2. Thank you so much! I've downloaded the images.

      I've just opened the images with Nikon Capture NX2. These targets are not just useful for comparing cameras! They can help decide which RAW converter does the best job.

      Right off the bat I can tell you that NX2 is so much better (with both cameras), it's not even close. Moiré is much less pronounced (both cameras). Resolution is quite improved.
      The resolution along the vertical scale is not the same as along the horizontal one. Maybe the target was not exactly perpendicular to the lens axis. These targets are so hard to shoot! Additionally on occasions two different sections of the image corresponding to the same spacing between bars show different results: maybe the target was not exactly planar.

      Comparing the cameras with NX2: there is very little chroma moiré with the D800 image. Moiré shows up in quite a few more places with the D800e, but it's still way less present and less pronounced than with the LR image. I've also mentioned that NX2 resolve much more detail. So if "6" corresponds indeed to the Nyquist limit there is quite a bit of false detail. But contrast between lines is quite improved right before the limit with NX2 as well.

      I personally can't make a call on resolution differences between the cameras, at least not without much more careful examination, because I think the target has minor issues as previously described.

      Delete
    3. Interesting observations you are making here. I'll check it out. Thanks.

      However, the observations you make are normal for test chart evaluations. The local performance is never consistent enough to derive stable conclusions by looking at the images. This is where a slanted edge measurement helps as it averages over hundreds of pixels. Therefore, I can tell that vertical and horizontal resolution are the same (at least with the LR PV2010 process) and that resolution slightly decreases when moving off center by even small distances (slightly but measurably). The chart is planar and parallel enough (using the mirror method). At least, it should be so ;)

      Even the preselection f the sharpest of the focus series would have been too difficult for the naked eye. Pixel peeping without measurement software is a pain. I always try to avoid it.

      Delete
  7. Thanks for doing this, Falk, but this link doesn't work for me....

    ReplyDelete
    Replies
    1. The link obviously is in some BNF variant. You need to remove {|} and select one of the two names to derive the valid URLs.

      Delete
  8. thanks for the great writeup, i just also blogged about your comparison and pointed my readers to your conclusions on the D800 vs the D800e. I just have a question regarding the two articles (english and german) -- the german writing is much more drastic in saying there is no real difference between the D800 and the D800e, while i am missing this part in the english text. has this just not been translated, or what is the reason?

    ReplyDelete
    Replies
    1. @nanofunk, sorry for the delay in responding. I need to check the spam blocker more often ;)

      The German writing is no exact translation as Dieter is co-author of the original article and participated in the work done. So, he may have a slightly different perspective on stuff. E.g., photography is his primary job while it is secondary for me.

      So, he has a more practical attitude "to get his stuff done". The level of differences found in our study is irrelevant to 99% of customers and I think this is what he wanted to express. The differences are measurable but hard to see in practice.

      Delete
  9. Nice review. Did you compare color in both models? the difference of Sharpness and Detail may not be relevant for most of us. Where tonality can still make a difference. I realized there was a bit difference between d700/d3 and d3x already. Now in some reviews you can see different colors on building and landscape. Also d800 compared to Hasselblad still shows that magenta, I think the d800e might show it less but I just depend on raw's that I found on the internet. this could be a reason to me to go for the "E" or even HB.

    ReplyDelete
    Replies
    1. @Anonymous, another post w/o a nick :(

      Technically, both D800 and D800E have the exact same color filters, only that one of them is turned 90° in one camera wrt the other. So at least for unpolarized light, there cannot be a difference in colors.

      Delete
  10. Thank you for your very interesting post. Do I understand correctly that with correct sharpening the d800 can resolve the same level of detail at the d800e? thanks

    ReplyDelete
    Replies
    1. Hi, I hoped to have explained that well enough. But obviously not ;) You are right for tack sharp images from the D800: they resolve down to the Nyquist frequency and can indeed be brought to the same level of detail resolution as images from the D800E. This is the typical situation in the studio.

      However, if the image from the D800 is already lacking in the finest detail then sharpening cannot bring it to the level of the D800E. And then color moiré won't be an issue either. This is the typical situation for action, wildlife, low depth of field portraiture or at low light.

      Delete
    2. ok thank you, so I understand that for landscape shots, there may be more detail in the shadows, but areas of already high contrast should sharpen to the same level of detail?

      Delete
    3. hello again, I just read the excellent link http://www.falklumo.com/lumolabs/articles/D800AA/D800AAFilter.html

      This is great.

      I had a couple of queries.

      1. regarding focusing control, it was not clear why the d800 is 12.8% less LWPH below the d800e.

      2. I didn't understand why the 4% difference was irrelevant, perhaps 115LP is not much in a picture but it's still 4%

      3. In the optimal resolution figure fig 6, I presume that 1px at 100% on the d800e gives suboptimal resolution?

      thanks for answering my tedious questions

      dhuksha

      Delete
    4. @dhuksha
      1. Well, the D800 has a Bayer-AA filter and we use the same sharpening parameters here. You may believe that an MTF50 figure expressed in LW/PH is a camera's resolution. It is not. It is the somewhat arbitrary spatial freqency where contrast becomes 50%.

      2. It IS irrelevant. Esp. as normalizing both frequencies at 50% contrast is somewhat arbitrary anyway. E.g., at 75% they are the same.

      3. Yes, it is oversharpened to the point where MTF50 actually decreases (as visible in comparison to 0.8px). The D800E under ideal image-taking circumstances should be sharpened with less than 1px radius.

      Delete
  11. Fantastic write up. LR cooks the data quite a bit before showing it to us, even with all sliders at zero. I wonder if to get more objective, uncooked MTF readings it wouldn't be better to use tools like dcraw.

    In the same vein, I wonder whether to measure the MTF of image data after having artificially stretched its acutance through fairly heavy use of USM may give somewhat arbitrary resuls. Perhaps sharpening through light deconvolution (Raw Therapee or Topaz Infocus) would be more appropriate for the type of measurements sought and give more objective results.

    ReplyDelete
  12. This is also a very interesting related article, in case you missed it:
    http://mtfmapper.blogspot.it/2012/06/nikon-d40-and-d7000-aa-filter-mtf.html

    ReplyDelete
    Replies
    1. Jack, thanks for the interesting link :)

      WRT measuring MTF based on RAW data... well, the data must either be demosaiced or the Bayer filter color responses must be calibrated for a given B&W target and single color source (which dcraw doesn't offer -- dxo may use that and regard as a trade secret). I consider MTF measurements based on demosaiced data to be ok, with demosaicing being part of the optical chain. Therefore, I prefer ACR (LR) over dcraw because it is a more realistic (or more widespread at least) effort at demosaicing.

      The same applies to sharpening. Yes, it introduces an arbitrary factor and I wrote extensively about it. Actually, I prefer USM over deconvolution because USM has its own MTF (it is a linear operator and its MTF exceeds 1) while LR-deconvolution has not.

      So, you are correct that one must take care to do a proper treatment of the effects you have mentioned.

      Delete
  13. Thank you Falk, having read through all this I keep substituting Pentax K5II for D800 and K5IIs for the E. I know the MP difference will argue against the Pentax but it would seem that the same conclusion would result or perhaps with a bias against the advantages of the IIs. As K5 owner I am after a second body and am wondering which way to jump.

    David McKeand

    ReplyDelete

Please if posting anonymously, choose a nickname for your post. Thanks.