Pages: [1]   Go Down

Author Topic: Diffraction-thoretical and real  (Read 5515 times)

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Diffraction-thoretical and real
« on: July 18, 2008, 02:26:39 pm »

I just posted these results on the DPReview D3 forum and decided to share them here.

Some of the theory about diffraction is posted on the Cambridge in Color web site.

According to Sean's analysis, when the Airy disc exceeds the pixel spacing, one can start to see degradation of resolution, and resolution is definitely degraded when the Airy disc is twice the pixel spacing. The pixel spacing of the D3 is 8.4 microns. The sizes of the Airy discs (green light, 530 nm) for f/2.8, 4, 5.6, 8, 11, 16, 22 and 32 are 3.6, 5.2, 7.2, 10.3, 14.2, 20.7, and 28.5, and 41.4 microns respectively.

I measured the MTF50 of the D3 with the Nikon 60 mm f/2.8 macro, which is one of the sharpest Nikon lenses, with Imatest. The results are without sharpening and are expressed in cycles per mm (line pairs/mm) at various apertures. For the test, I used raw files and Adobe Camera Raw for conversion. Resolution in the horizontal and vertical axis was the same (unlike with the D200 and some other Nikon cameras). I used auto-focus and three different focusing attempts. The results are the average of 6 values.

[attachment=7512:attachment]

F/4 through f/11 show near maximal resolution, but there is some degradation wide open and at f/16. At f/32 there is severe degradation.

Here is an Imatest screen capture for f/5.6. Please refer to the Imatest web site for details of the methodology. These results should be similar to what one would get with the Canon 5D. One might think these resolutions are low. The MTF50 at f/5.6 with standardized sharpening was 40.8 lp/mm with the D3 and the result for the Canon 5D with an image downloaded from DPreview.com was 44.4 lp/mm. I then downloaded an image for the D3 and obtained an MTF50 of 39.9. The DPreview Nikon image was undersharpened by 14.1% and the Canon image oversharpened by 10.8% according to the Imatest analysis.

[attachment=7513:attachment]

One can look at many such tests for various cameras and lenses (mainly cropped frame dSLRs at PhotoZone.de. With these cameras, the best resolution is often at f/5.6 with good primes.

Bill
« Last Edit: July 19, 2008, 08:04:19 pm by bjanes »
Logged

larkis

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 332
    • My photography blog
Diffraction-thoretical and real
« Reply #1 on: July 18, 2008, 11:33:52 pm »

How would this apply to 4x5 film ? Would f32 loose just as much sharpness as in 35mm or would the large area of the film help with that ?

Tony Beach

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 452
    • http://imageevent.com/tonybeach/twelveimages
Diffraction-thoretical and real
« Reply #2 on: July 19, 2008, 02:50:56 am »

I recently did my own informal test of this issue.  The diffraction of the D70 (not to be confused with the D700) and D300 highlight the issue nicely:



There is no drop-off on the D70 going from f/11 to f/16 while the drop-off between those apertures for the D300 is obvious.  However, the D300 at f/16 is just as sharp (presuming equal enlargement) as the D70.  The lesson is that if you buy a high MP DSLR and stop it down, you end up with the same results you would have gotten from a less expensive DSLR.  The solution when you reach the diffraction limitations is to use tilt/shift lenses to attain greater DOF.  This has convinced me that a 24 MP DSLR stopped down past about f/11 is probably no better than a 12 MP DSLR stopped down to f/16; and for deep DOF landscape photography it is wiser to first buy some tilt/shift lenses before buying a high MP DSLR.
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Diffraction-thoretical and real
« Reply #3 on: July 19, 2008, 04:26:02 am »

These results are roughly consistent with my observations. One should bear in mind also that the resolution differences between F8 and F16 with a camera such as the D3 or 5D will likely be less significant using a medium quality zoom which will likely not be quite as sharp at F8 as a high quality prime, but probably equally as sharp at F16. All lenses tend to be equal at F16.
Logged

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Diffraction-thoretical and real
« Reply #4 on: July 19, 2008, 09:44:15 am »

Quote
How would this apply to 4x5 film ? Would f32 loose just as much sharpness as in 35mm or would the large area of the film help with that ?
[a href=\"index.php?act=findpost&pid=209319\"][{POST_SNAPBACK}][/a]

Stopping down has the same effect on resolution with both 35mm and 4x5 inch formats. However, the net effect of stopping down will be less with the 4x5 format, since the image will need less magnification to print at a given size. For example, an 8 by 10 inch print requires only 2x with 4x5 but 10x with 35 mm.

Bill
Logged

lbalbinot

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 70
    • http://luisbalbinot.com/
Diffraction-thoretical and real
« Reply #5 on: July 19, 2008, 10:18:44 am »

Very good analysis bjanes! I'm glad to know because f/11 is my main aperture setting. I also liked the analysis of the sharpness setting of the Canon 5D and Nikon D3. Some people that only shoot JPEGs keep saying that the 5D is sharper for this and that reason. I always knew that 5D was doing a more aggressive sharpening.

Thanks for sharing your results!

Luis
Logged
Luis F Balbinot
luisbalbinot.com

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Diffraction-thoretical and real
« Reply #6 on: July 19, 2008, 12:26:57 pm »

Hi,

Physics apply to all formates. Film flatness is definitively an issue with 4x5 and large format lenses are generally not well corrected at maximum aperture, except the Rodenstock and Schneider "digital lenses" which do not cover 4x5.

So the answer is both yes and no, optimal sharpness on MF lenses is probably around f/11. With f/32 there would definitively be a loss of sharpness. You would probably not see this loss in an enlarger but certainly in a very high resolution scan like at 3200 PPI. A print from a 3200 PPI scan at 200 DPI would be 64"x80", pretty big by any means.

Lars Kjellberg who had the original "Photodo" site actually made an experiment with 135, 120 and 9x12cm film where he compared pictures taken at apertures f/5.6 (135), f/11 (120) and f/22 (9x12cm). Sharpness was essentially the same, when using T-max 100 film.

Check article here: http://old.photodo.com/art/35_m9.shtml

A remark on the side. I made a 70x100 cm print from a scanned Velvia. Scanning was done at 3200 PPI. At actual pixels the scan was pretty ugly, but I applied all "the usual stuff" on it. Grain reduction, sharpening with edge masking and so on. The print was incredibly sharp at the point of focus, the background was not sharp at all but good enough so no one would react. Sharpness "rolled off" very smoothly. This was another example that a print is different from what you see when pixel peeping at actual pixels.

I made a similar test a while ago with an 80-200/2.8 Minolta APO lens on two different cameras. The result was pretty much similar:

http://luminous-landscape.com/forum/index....pe=post&id=3387

I also enclose an example of same picture taken with f/11 and f/32.

Best regards
Erik



Quote
How would this apply to 4x5 film ? Would f32 loose just as much sharpness as in 35mm or would the large area of the film help with that ?
[a href=\"index.php?act=findpost&pid=209319\"][{POST_SNAPBACK}][/a]
« Last Edit: July 19, 2008, 01:18:40 pm by ErikKaffehr »
Logged
Erik Kaffehr
 

ejmartin

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 575
Diffraction-thoretical and real
« Reply #7 on: July 19, 2008, 03:03:24 pm »

Quote
According to Sean's analysis, when the Airy disc exceeds the pixel spacing, one can start to see degradation of resolution, and resolution is definitely degraded when the Airy disc is twice the pixel spacing. The pixel spacing of the D3 is 8.4 microns. The sizes of the Airy discs (green light, 530 nm) for f/2.8, 4, 5.6, 8, 11, 16, 22 and 32 are 2.6, 3.6, 5.2, 7.2, 10.3, 14.2, 20.7, and 28.5, and 41.4 microns respectively.

I think you may want to fix this, there are 8 apertures and 9 Airy disk sizes listed (I think you started your list of Airy disk sizes at f2).

Quote
I measured the MTF50 of the D3 with the Nikon 60 mm f/2.8 macro, which is one of the sharpest Nikon lenses, with Imatest. The results are without sharpening and are expressed in cycles per mm (line pairs/mm) at various apertures. For the test, I used raw files and Adobe Camera Raw for conversion. Resolution in the horizontal and vertical axis was the same (unlike with the D200 and some other Nikon cameras). I used auto-focus and three different focusing attempts. The results are the average of 6 values.

[attachment=7512:attachment]

F/4 through f/11 show near maximal resolution, but there is some degradation wide open and at f/16. At f/32 there is severe degradation.

Bill
[a href=\"index.php?act=findpost&pid=209200\"][{POST_SNAPBACK}][/a]

It looks as though diffraction is setting in somewhere between f11 and f16, ie somewhere above an Airy disk size of 14.2µ according to your figures.  Now, to compare that with theory.  The theoretical resolution of the Bayer array is sqrt[2] times the pixel spacing (12µ for the D3), since that is the resolution of the Green channel sensels which are diagonally spaced in the array.  The AA filter also takes a bite out of the resolution and may degrade that figure a bit more.  Raw converters may not deliver all that resolution either, however, depending on their algorithms; for that reason it would be nice to be able to do resolution tests on the raw data directly, though I've not seen that done (I did something crude to use raw data for resolution tests on the 40D and D300, but it was not as systematic or as accurate as I would like).  I suspect one can attribute the 2µ difference between the 12µ resolution of the Bayer array and the 14µ or a bit more Airy disk size at the onset of diffraction effects, to the combination of AA filter and conversion software.
« Last Edit: July 19, 2008, 03:04:15 pm by ejmartin »
Logged
emil

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Diffraction-thoretical and real
« Reply #8 on: July 19, 2008, 09:22:16 pm »

Quote
So the answer is both yes and no, optimal sharpness on MF lenses is probably around f/11. With f/32 there would definitively be a loss of sharpness. You would probably not see this loss in an enlarger but certainly in a very high resolution scan like at 3200 PPI. A print from a 3200 PPI scan at 200 DPI would be 64"x80", pretty big by any means.

Lars Kjellberg who had the original "Photodo" site actually made an experiment with 135, 120 and 9x12cm film where he compared pictures taken at apertures f/5.6 (135), f/11 (120) and f/22 (9x12cm). Sharpness was essentially the same, when using T-max 100 film.

Check article here: http://old.photodo.com/art/35_m9.shtml

[a href=\"index.php?act=findpost&pid=209378\"][{POST_SNAPBACK}][/a]

I remember that article well. It was an eye-opener for me when I first read it several years ago, before I'd bought my first digital camera.

There are a few points worth stressing here.

(1) Large format lenses generally don't show any noticeable resolution fall-off until they are stopped down beyond F22.

(2) When comparing resolution in different formats, it's only sensible to make F stop adjustments to equalise DoF. Is there anyone composing pictures or artworks who takes the attitude, 'I'm only concerned with resolution in the plane of focus. I don't care about the foreground or background'?

However, I suppose there are circumstances where such an attitude might be appropriate; if the purpose was to create a montage from various cut-outs, for example, or to glean the maximum amount of detail from part of a scene for scientific purposes.

(3) The film used in Lars Kjelberg's tests, that demonstrated the 35mm shots at F5.6 were as sharp as the 4x5" shots at F22, is a film with remarkable MTF properties.

T-Max 100 can apparently record up to 50 lp/mm without any loss of contrast, ie. MTF response at 50 lp/mm is still 100%. By comparison, cameras such as the 1Ds, the 5D and the Nikon D3 are approaching zero MTF at 50 lp/mm.

Even at 100 lp/mm, a resolution which no digital camera is capable of achieving (except perhaps certain scanning backs and small P&S cameras), T-Max 100 still has an MTF response of about 65%.

I believe this is why the 'Film versus Digital' debate keeps occasionally resurfacing. The potential is there for 35mm film to outresolve any current DSLR (including the 1Ds3), but only at great expense and inconvenience, and probably only with certain B&W film types, many of which have now been discontinued.

When Michael, some years ago, tried to demonstrate that the 1Ds was roughly the equivalent of 6x7cm film format, in terms of detail captured, he used the popular Fuji Velvia film. The MTF response of Velvia at 100 lp/mm is a mere 15%.

And, of course, when trying to extract the maximum detail from film, one needs a very expensive, high resolution drum scanner. My highest resolving scanner is the Minolta Dimage 5400 II. I've seen tests on the internet which suggest that this scanner has a maximum resolving capability of around 70 lp/mm from B&W film, and less from color film.
« Last Edit: July 19, 2008, 09:26:21 pm by Ray »
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Diffraction-thoretical and real
« Reply #9 on: July 20, 2008, 01:40:40 am »

Thanks for thoughtful comments!

Erik

Quote
I remember that article well. It was an eye-opener for me when I first read it several years ago, before I'd bought my first digital camera.

There are a few points worth stressing here.

(1) Large format lenses generally don't show any noticeable resolution fall-off until they are stopped down beyond F22.

(2) When comparing resolution in different formats, it's only sensible to make F stop adjustments to equalise DoF. Is there anyone composing pictures or artworks who takes the attitude, 'I'm only concerned with resolution in the plane of focus. I don't care about the foreground or background'?

However, I suppose there are circumstances where such an attitude might be appropriate; if the purpose was to create a montage from various cut-outs, for example, or to glean the maximum amount of detail from part of a scene for scientific purposes.

(3) The film used in Lars Kjelberg's tests, that demonstrated the 35mm shots at F5.6 were as sharp as the 4x5" shots at F22, is a film with remarkable MTF properties.

T-Max 100 can apparently record up to 50 lp/mm without any loss of contrast, ie. MTF response at 50 lp/mm is still 100%. By comparison, cameras such as the 1Ds, the 5D and the Nikon D3 are approaching zero MTF at 50 lp/mm.

Even at 100 lp/mm, a resolution which no digital camera is capable of achieving (except perhaps certain scanning backs and small P&S cameras), T-Max 100 still has an MTF response of about 65%.

I believe this is why the 'Film versus Digital' debate keeps occasionally resurfacing. The potential is there for 35mm film to outresolve any current DSLR (including the 1Ds3), but only at great expense and inconvenience, and probably only with certain B&W film types, many of which have now been discontinued.

When Michael, some years ago, tried to demonstrate that the 1Ds was roughly the equivalent of 6x7cm film format, in terms of detail captured, he used the popular Fuji Velvia film. The MTF response of Velvia at 100 lp/mm is a mere 15%.

And, of course, when trying to extract the maximum detail from film, one needs a very expensive, high resolution drum scanner. My highest resolving scanner is the Minolta Dimage 5400 II. I've seen tests on the internet which suggest that this scanner has a maximum resolving capability of around 70 lp/mm from B&W film, and less from color film.
[a href=\"index.php?act=findpost&pid=209470\"][{POST_SNAPBACK}][/a]
« Last Edit: July 20, 2008, 01:46:17 am by ErikKaffehr »
Logged
Erik Kaffehr
 

juicy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 254
Diffraction-thoretical and real
« Reply #10 on: July 20, 2008, 08:12:51 am »

Hi

My completely unscientific tests have shown that with the original 1Ds and Canon 100mm macro the most detail in the plain of focus is captured @ f5.6 or f6.3. @ f16 the smallest details have disappeared. In most cases and in smaller prints f11-16 is the best compromise when more dof is needed. Depending on the subject even f28 may produce good results when the print size is less than A4 (ca 8x12").

Cheers,
J
Logged

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Diffraction-thoretical and real
« Reply #11 on: July 20, 2008, 08:24:22 am »

Quote
I think you may want to fix this, there are 8 apertures and 9 Airy disk sizes listed (I think you started your list of Airy disk sizes at f2).
[{POST_SNAPBACK}][/a]

Thanks, I took care of that error.

Quote
It looks as though diffraction is setting in somewhere between f11 and f16, ie somewhere above an Airy disk size of 14.2µ according to your figures.  Now, to compare that with theory.  The theoretical resolution of the Bayer array is sqrt[2] times the pixel spacing (12µ for the D3), since that is the resolution of the Green channel sensels which are diagonally spaced in the array.  The AA filter also takes a bite out of the resolution and may degrade that figure a bit more.  Raw converters may not deliver all that resolution either, however, depending on their algorithms; for that reason it would be nice to be able to do resolution tests on the raw data directly, though I've not seen that done (I did something crude to use raw data for resolution tests on the 40D and D300, but it was not as systematic or as accurate as I would like).  I suspect one can attribute the 2µ difference between the 12µ resolution of the Bayer array and the 14µ or a bit more Airy disk size at the onset of diffraction effects, to the combination of AA filter and conversion software.
[a href=\"index.php?act=findpost&pid=209409\"][{POST_SNAPBACK}][/a]

Emil,

A very insightful analysis, as usual. However, I have a slightly different take on the matter and would appreciate your input. If your resolution test grid has alternating black and white lines and the Bayer array pixels are aligned parallel with the grid, all of the sensels can be used and the resolution of the sensor (Nyquist limit) is equal to the pixel spacing. As you point out, the demosaicing algorithm and the blur filter decrease resolution. The ideal Bayer array would have an MTF 50 of 0.5 cycles per pixel, but [a href=\"http://www.imatest.com/docs/tour_sfr.html]Norman Koren[/url] has found that 0.33 cycles/pixel is very good performance in actual cameras.

This graph summarizes my results. The measured system MTF 50 at various apertures is shown along with the MTF 50 of an ideal lens, the Nyquist limit of the camera, and Nyquist / 1.414 as you suggested. We see that the MTF 50 of an ideal lens equals the Nyquist frequency between f/11 and f/16. Some would say that the system is "diffraction limited" at that aperture.

Maximal resolution is at about f/5.6, when stopping down from the maximum aperture of f/2.8 reduces lens aberrations. Beyond this point, stopping down further decreases resolution, and I would say that this is due to diffraction.

[attachment=7549:attachment]

Ay you know, system MTF does not equal the weakest link in the imaging system, since system MTF is obtained by multiplying the MTFs of the components. A high MTF in one component can partially offset a low MTF in another component. The MTF 50 of the Bayer array is not easily obtained, but Norman Koren gives an approximation here. The MTF 50 of an ideal lens at f/5.6 is much greater than the MTF of the sensor, and the MTF of the actual lens is also most likely considerably greater than that of the sensor. This high lens MTF improves system MTF. As one stops down further, lens MTF of the lens decreases and brings down the system MTF, even though the lens is still out resolving the sensor.

Bill
« Last Edit: July 20, 2008, 08:29:32 am by bjanes »
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Diffraction-thoretical and real
« Reply #12 on: July 20, 2008, 11:06:16 pm »

Quote
Ay you know, system MTF does not equal the weakest link in the imaging system, since system MTF is obtained by multiplying the MTFs of the components. A high MTF in one component can partially offset a low MTF in another component. The MTF 50 of the Bayer array is not easily obtained, but Norman Koren gives an approximation here. The MTF 50 of an ideal lens at f/5.6 is much greater than the MTF of the sensor, and the MTF of the actual lens is also most likely considerably greater than that of the sensor. This high lens MTF improves system MTF. As one stops down further, lens MTF of the lens decreases and brings down the system MTF, even though the lens is still out resolving the sensor.

Bill
[a href=\"index.php?act=findpost&pid=209520\"][{POST_SNAPBACK}][/a]

Bill,
We should also bear in mind that precise information on the MTF properties of the components in the system is generally not available. If it were, we could probably dispense with all paractical testing and deduce expected performance from the specifications alone.

Generally, whenever I've been disappointed in the sharpness of an image I've taken, it's because I've made a mistake along the lines of not using a fast enough shutter speed for a hand-held shot, not focussing accurately enough, neglecting to enable MLU, forgetting to disable IS when using a tripod, or forgetting to enable IS when the tripod is a bit wobbly and subject to the instability of a slight breeze, miscalculating the appropriate F stop for a desired DoF.

Getting technique right is often more important than possession of high quality equipment. To achieve the greater sharpness that expensive lenses are usually capable of, often requires a more rigorous technique.

The 1/FL rule for shutter speed cannot apply. Rather, it should apply in conjunction with IS. With a 400mm lens, for tack sharp images one should perhaps be using 1/400th sec exposure with IS enabled, rather than 100th sec because IS provides a 2 or 3 stop advantage.
Logged

ejmartin

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 575
Diffraction-thoretical and real
« Reply #13 on: July 25, 2008, 09:25:48 am »

Quote
Emil,

A very insightful analysis, as usual. However, I have a slightly different take on the matter and would appreciate your input. If your resolution test grid has alternating black and white lines and the Bayer array pixels are aligned parallel with the grid, all of the sensels can be used and the resolution of the sensor (Nyquist limit) is equal to the pixel spacing. As you point out, the demosaicing algorithm and the blur filter decrease resolution. The ideal Bayer array would have an MTF 50 of 0.5 cycles per pixel, but Norman Koren has found that 0.33 cycles/pixel is very good performance in actual cameras.

Bill
[a href=\"index.php?act=findpost&pid=209520\"][{POST_SNAPBACK}][/a]

Yes you're right.  For resolving lines rather than points, the limits are very much dependent on the orientation of the lines relative to the pixel grid.  I was talking about point resolution, but of course there is much detail in images that comes from lines.
Logged
emil
Pages: [1]   Go Up