The distance between two pixels of the same color can never be "1 pixel width", since they do not populate every pixel instance. The absolute minimum distance between two green pixels is 1.41 x pixel size, sqrt(2) since the distance is a diagonal 45º distance.
Thanks for the insightful comments theSuede. Does the distance calculation (diagonal as opposed to pixel width) really apply for white light? I can see image degradation with stopping down on my cameras (FF with 18-21MP) closer to the calculation that uses pixel width. I'd say it is somewhere in between. 1 and sqrt(2)...
Toothwalker wrote:
That is true, but in terms of resolution a lens cannot be better than a diffraction-limited lens. Thus ... how can this lens be so much better than the supposedly (close to) diffraction-limited Canon lens?
The Distagon does have a smaller image circle than the TS-E.
I shot a Siemens star with the Distagon at f/2.8 at distance 8m. Once again it blew my mind. It produced aliasing on the Q sensor. Limiting resolution before aliasing set in was about 248 cy/mm.
Edit: The TS-E isn't truly diffraction limited at full aperture, I only said it was close to in the test on the Canon board. It does lose more contrast than the distagon at smaller apertures though.
Mirek Elsner wrote:
Thanks for the insightful comments theSuede. Does the distance calculation (diagonal as opposed to pixel width) really apply for white light? I can see image degradation with stopping down on my cameras (FF with 18-21MP) closer to the calculation that uses pixel width. I'd say it is somewhere in between. 1 and sqrt(2)...
White light is green, as seen from the PoV of a Bayer based sensor. Green has the highest spatial resolution. You can't even MEASURE the strict diffraction loss in the red and blue channels in the raw file until you are at almost 2x pixel width Rayleigh (which is also the distance between the closest pairs of blue and red pixels, 2p c-c)...
Something smaller (in diffraction) than this can indeed be SEEN in an interpolated, finished imaged as a drop in p-p contrast - but the question you should ask yourself is WHAT you are seeing. Signal theory does not lie, and can very rarely be cheated. What you're seeing as "more detail and sharpness" is made up to a very large portion of aliasing and false detail created by the interpolation engine in the raw converter. Yes, it can look "nice" - but it has very little to do with the actual reality in front of the lens.
Toothwalker wrote:
That is true, but in terms of resolution a lens cannot be better than a diffraction-limited lens. Thus ... how can this lens be so much better than the supposedly (close to) diffraction-limited Canon lens?
The total Strehl is no doubt a lot worse in the large Canon lens, even if it's very, very close to diffraction limit when inspected at the intended projection scale...
Most better phone camera module lenses (average cost for the wholde module incl. PIC ~7-18 USD)can outresolve a Leica 100 APO-elmarit by a factor of three or more, if you just look at the lp/mm contrast. But in a much smaller projection circle...
The lens in the newest Nokia is more than twice as good at 160lp/mm as the very best photographical lens that Leica has ever produced. This is possible when you plan the optics for interferometry, a luxury you cannot afford in large-system lenses (it's plain physically impossible to keep the tolerances needed - sometimes down to quarter-wavelength gapwidths in the laminations...).
*Interference and diffraction are in practice the same thing, at least we cannot find any conclusive differences between the phenomena.
So yes, you can do "better than diffraction limited".
Thanks for doing this test. I had read that the theoretical "diffraction limit" on my 7d is f6.7. I will worry less about loss of quality due to diffraction after looking at your tests.
I did some of my own tests a while back using my 5d2 and a sharp lens (Leica 100 2.8 apo). I did macro shots of coins at f5.6 8 11 & 16. I could detect some loss of sharpness at f16, hard to tell a difference at f11.
I would be interested in testing a 5d2 and 7d with a set of equal distance shots and all apertures. I wonder if more detail or sharpness would be found on a f16 7d shot than the best 5d2 shot ?
jpeter wrote:
I would be interested in testing a 5d2 and 7d with a set of equal distance shots and all apertures. I wonder if more detail or sharpness would be found on a f16 7d shot than the best 5d2 shot ?
Thanks again, JP
JP, In my experience the 7D does not resolve more detail than the 5DII at f/16, but I haven't done any test with very fine color detail.
Here is a sample with a Siemens star with diameter 110 mm. The test target printed at 720 ppi doesn't have any problem with true detail until the darker part towards the middle.
Unscaled pixel crop from the Pentax Q, no sharpening applied. Zeiss Distagon 2.8/21 ZE aperture f/3.5
Toothwalker wrote:
That is true, but in terms of resolution a lens cannot be better than a diffraction-limited lens. Thus ... how can this lens be so much better than the supposedly (close to) diffraction-limited Canon lens?
theSuede wrote:
The total Strehl is no doubt a lot worse in the large Canon lens, even if it's very, very close to diffraction limit when inspected at the intended projection scale...
.
It turned out to be a flare issue with the TS-E 24 II:
I have been pleasantly surprised how well the Zeiss 35/2 and 21/2.8 have performed across the frame on the D800E. I still hope Canon gets its sensor act together but they are now way behind Sony sensor tech in D800(E) and financially it makes sense to probably can any expensive low volume sensor design/research/fab which is at. Volume/price disadvantage vs the Sony FF sensors.
carlitos wrote:
In a 24x36mm FF application, what is the optimum MP size, considering DOF and diffraction? Seems the discussion so far is close to stating it.
What do you mean by "optimum considering DOF and diffraction"?
That's an easy question, but it has no fixed answer. It depends on your DoF needs... :-)
The linear resolution needs to be about 1.5x the diffraction limit at your intended DoF. At that point you don't need an AA-filter, you can have less coverglass thickness, you utilize the red and blue pixels to the max - and so on. This has a "maximum linear resolution" cap at the current technological limit of course.
There's always an efficiency decrease at some level, right now that seems to be at about 1.1µm for small-format sensors and about 3.5µm for larger format sensors. When you make the pixels smaller than that you loose efficiency, get more noise, and increase the losses at very large apertures.
You can also set this in relation to your image resolution need: If you need 8MP in you end use, then the ideal resolution is about 2-3x that MP amount. 16MP will be enough in many cases, but about 25 is the "safe limit" where you utilize very close to 100% information access in each and every pixel in the finished 8MP result. This is something you just CANNOT achieve when you try to use 100% scale straight from the sensor.
I was forgetting about end use or purpose of the final image. But what about, let's say, a 60" "Retina" screen in a typical 10'(3m) x 12'(4m) room. Would the optimum number of pixels vary with the DOF?
I mean, it does seem that there is a maximum number of pixels that 24mmx36mm is useful for, depending on the purpose of the image.
