My definitive D800E vs Canon 5D Mark III image sharpness comparison, in DAP.
The usual Retina-grade images and crops are presented.
A 27% linear resolution difference is modest (D800E vs 5DM3). It is 22 vs 36 megapixels, or 12 vs 19, etc.
A 27% linear resolving power difference is readily evident if the lens and and sensor cooperate optimally, but it doesn’t “jump out” necessarily. And with a sub-par lens, or less than absolutely perfect technique, the differences quickly evaporate.
The genesis of this effort was the DxO report I discussed, which claimed two things: (1) that at the same resolution the Canon 5D Mark III showed 15% more resolution than the 5D Mark II*, and (2) that the Nikon D800 was not necessarily delivering its rated resolution by comparison (“In terms of pixel count and stills output, 36 Mpix to 22.3 Mpix sounds a lot but in real life conditions, it’s not as much as the figures suggest”).
So I wanted to investigate this claim with respect to the D800E and 5D Mark III using the best lens and most meticulous effort possible.
Accordingly, I chose the D800E and Canon 5D Mark III along with the Zeiss 135mm f/2 APO-Sonnar for both.
* Which strongly suggests that the sensor or glass over the sensor has to be responsible, hence the Nikon designs would possibly show similar limitations, and the new Nikon D7100 in fact seems to perform better than one would expect compared to its D800E sibling.
Paraphrasing some comments:
- “You should also compare the D800 to D800E and the D800 to the 5DM3”.
- “My old 20mm Nikon lens is not retrofocus and it behaves differently on different Nikon cameras, therefore you should test with other lenses”.
- “You should not sharpen the images” or “there are sharpening artifacts” or “I don’t like your sharpening because at 200% I can see sharpening artifacts”.
- “That staircasing effect on the Canon image must be a fault of your sharpening”.
I work in the real world for photographs, which means I produce a result which I deem suitable for presentation and/or print. Not some unsharpened variant that looks like a blur filter was run on it—which in fact is blurry as a result of discrete sampling and demosaicing, lacking acutance that must be restored as a natural part of working properly with digital. And that staircasing on the Canon image is present with zero sharpening, it’s just harder to see with no acutance.
I don’t claim to have all the answers to every combination of lens and camera, but I did this test with exceptional precision, taking two full days to do this one comparison properly (2nd day was a repeat confirmation). A 4-way test becomes far more demanding and time consuming, and raises the odds for errors of all kinds.
The Zeiss 135mm f/2 APO-Sonnar has an exceptionally digital-friendly chief ray angle, and I would call it out as the best lens available on a DSLR today (see How Does the Zeiss 135mm f/2 APO-Sonnar Perform on a 56-Megapixel DSLR?). The 20mm comment is out in left field in this context, being retrofocus in nature as any 20mm DSLR lens has to be in order to clear the mirrorbox.
Older wide angle designs also had fairly steep ray angles (being designed for film) compared to modern digital designs like the Zeiss ZF.2 15/2.8 Distagon, which remarkably has a ray angle even more friendly to digital than the 100/2 Makro-Planar.
As far as I know, there is no better lens for this test than the 135/2 APO-Sonnar in favoring the D800E (and 5DM3) for peak quality, which was my hypothesis to prove or disprove: to establish a clear image sharpness advantage with everything being its best.
The results are what they are, the only issue is what is the cause, my theory being discussed in the coverage. I’ve made certain inquiries of experts, and perhaps they can shed some light on the results.
Massimo T writes:
I have a theory I wanted to run by you.
I have noticed over the years, companies such as canon have been "baking" Raw data. I thought this was limited to noise reduction but would it be possible for them to also sharpen or apply extra contrast before storing the Raw data?
On another note, looking at DXOmark lens data, the D800 (non E) performs at the same "MP" rating as the D600 and D3x. They only tested three lenses on the D800E but it does seem to have higher performance than the other full frames but not publishing more results for the D800e also makes me think they ran into the same issue as you and as a favor to Nikon (and their reputation) they are hesitant to publish more results.
A can of worms just been opened but the camera is still as good as it was a month ago.
DIGLLOYD: Electronics and sensor being what they are, some processing might be necessary and proper given the nature of the medium. And I expect higher density sensors will do this silently to help with diffraction at the hardware level, if they are not already doing so.
At present, it is my assumption that Nikon does some black-level processing and this is why it is so cleaner at near-black compared to Canon. But it might also be sensor performance (probably both), and I don't have any inside knowledge here.
As far as I know, Canon has long been favored by the astronomy photographers because it allows image stacking to eliminate noise; this cannot be done if the camera is processing the low order bits to suppress noise (and faint stars).
T. B writes:
One plausible hypothesis could be that the larger sharpness plane in the Z-axis (DOF) on Canon, could be a result of a combination of 1) blurring of the signal in the Low pass/AA filter, and 2) more aggressive in-camera sharpening algorithm after the AA blurring, resulting in a more widespread sharpness-plane in the 5D III compared to the D8000E.
It’s just speculations, but one never knows. Maybe you can try comparing with all sharpening routines, in the cameras (both) and in software, switched off?
DIGLLOYD: There has to be some reasonable cause to speculate and form a hypothesis. I do not deem this hypothesis reasonable. I shoot best quality raw only (max bits, lossless compressed), and it would be a terrific story to show that JPEG sharpening settings are actually sharpening the raw file with the Canon 5D Mark III, and that everyone on the planet has not noticed this. But this does not rule out silent electronic acutance enhancement in converting analog to digital.
Sensor designs including the cover glass can have significant effects and as per DxO, changes were made with the 5D Mark III yielding 15% better resolving power over the 5D Mark II (but this is not a claim I have verified).
Very interesting to see how unreliable pure pixel count may be to conclude about perceived image quality. The explanation hypothesis is of course quite speculative. We don't know exactly the filter design in both cameras, we also don't know what both parties mean with RAW -it might also be pre-cooked a bit.
I think I remember that Canon had announced with the MkIII something like deconvolution image restoration, but I also don't know exactly.
Would be nice to have time to dig deeper in these things - yet the needs of practical picture taking with lenses are a bit distant from that.
DIGLLOYD: It is very hard to say what is the cause. But one thing stands out to me: the Nikon D7100 with its much higher pixel density does not seem to exhibit the D800E behavior. The D7100 has no optical low pass filter, though Nikon does not go into details on exactly what glass is used on the sensor. See Zeiss 50mm f/2 Makro-Planar: Future Potential on ~56 Megapixel DSLR and How Does the Zeiss 135mm f/2 APO-Sonnar Perform on a 56-Megapixel DSLR?.
On the subject of “cooking” the raw file: it seems likely that Nikon does this because the dark tones are so clear in the near-black. Canon DSLRs have long been favored by astrophotographers because faint stars which might look like noise can be extracted by stacking 100 frames or more, something not possible with a camera raw that smooths away near-black pixels.
As for “cooking” the raw file with deconvolution, from what I see with many small sensor cameras I would be surprised if this is not already being done, and it’s quite possible that the Canon 5D Mark III employs such an approach also.