Is the problem of diffraction over-rated?

[ErikKaffehr]

Hi,

I would say that there is always a risk that sharpening to much may create artifacts instead of real detail.

Look at the images on this link: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/49-dof-in-digital-pictures?start=2

You can see that the sharpened images are very crisp, but they lack "finesse" in form of fine detail.

Another example is here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/68-effects-of-diffraction (end of page).
The fine detail within the read square is lost, although sharpening is extensive.

Best regards
Erik

But it seems maybe you are "assuming" you would lose it anyway when it goes to print so it doesn't matter  .  I guess I'm not sure that's always the case.  Sometimes you may lose it and it doesn't matter, but it seems there certainly may be times it would improve the image if it was preserved and it might be more visible than one might think.

I think we're basically on the same page ... just a different perspective on it.

[Walt Roycraft]

Erik, I just wanted say thanks for your information. Very helpful and much appreciated.

[Wayne Fox]

But otherwise if I want DOF f/22 will do fine as long as it doesn't drive my shutter speed down to around the half second area where the "thump" of the shutter on the 5DII shows up worst.

Not sure if you are aware but the 1st curtain of the 5DII is not mechanical but electronic.  This means if you use live view in place of mirror up, there are no moving parts in the exposure process until the end of the exposure when the second curtain releases.

[JohnCox123]

I want to point out that with an optimally sharpened image you won't notice diffraction. Every image should be sharpened optimally.

[David Sutton]

Not sure if you are aware but the 1st curtain of the 5DII is not mechanical but electronic.  This means if you use live view in place of mirror up, there are no moving parts in the exposure process until the end of the exposure when the second curtain releases.

No I didn't know that. I realise I was thinking of my 40D which wouldn't give a sharp image at around half a second even with mirror lock-up. With the noise the 5DII makes I assumed it would be the same!

[Fine_Art]

Here is the resolution of the D600 with 85 1.8 ISO100 no sharpening.

The way this siemens star chart works for those not familiar with it is a circle of 92 pixels is the angular equivalent of the linear nyquist limit of 2 pixels to resolve a line. The 92 is from Bart's pick of 144 cycles around. I am not painting the circles in for you because the re-jpg would squash some detail.

[xpatUSA]

Here is the resolution of the D600 with 85 1.8 ISO100 no sharpening.

Interesting images, thank you for taking the time to show them to us.

You were smart to darken the central area where printer moire can appear. I tried to post a series of the same target to demonstrate a lens "sweet spot" but forgot to smudge out the middle bit. Therefore, it all got very confusing when the discussion veered sadly off-course into the area of sensor resolution.

[Fine_Art]

Interesting images, thank you for taking the time to show them to us.

You were smart to darken the central area where printer moire can appear. I tried to post a series of the same target to demonstrate a lens "sweet spot" but forgot to smudge out the middle bit. Therefore, it all got very confusing when the discussion veered sadly off-course into the area of sensor resolution.

Yeah, I had to re-do it from before. When I first looked at the images I thought the pattern I was getting was from the camera/lens. When I took the target down I noticed the same pattern from the laser printer. I think it was inside the 92 pixel circle so it was rubish but I felt it might give people the wrong impression. I jifffy markered out past the strangeness to make very sure anything anyone saw in the image was from the camera/lens.

I did the test to make sure I want to keep this system. I might as well post it for others to see. It may be my imagination, or the weak low pass filter on the sonys but diffraction seems to jump in fast on the APS-C sonys. I almost never go past f11. On this setup I really cant see much difference going from 1.8 to f16 and anywhere in between. The contrast certaily goes up f5.6-f8. The actual finest detail seems to be about the same. The system has pixels to spare with monster high ISO capability which I have never had before so it is a keeper. FF D600 for wide angles and A55 for telephoto. Works for me.  I like my minolta/sony primes. DxO is on crack for the low ratings they give those lenses. This nikon lens is very nice but it is not a big gulf from the Sony macros. And it is easy to get tack sharp pictures with the SLTs due to no mirror slap. With a regular camera you need to put up with the MLU delay to get similar. Think about getting 10fps with MLU...

[Bart_van_der_Wolf]

You were smart to darken the central area where printer moire can appear. I tried to post a series of the same target to demonstrate a lens "sweet spot" but forgot to smudge out the middle bit. Therefore, it all got very confusing when the discussion veered sadly off-course into the area of sensor resolution.

Hi Ted,

Discussions can go off-topic when people don't understand what they are looking at. The target itself is already a challenge for the printer (and can be used for that as well, using a loupe and calipers to measure the blur diameter), but that also gives information about whether the print is offering reasonably more detail than the lens can resolve, and how the Raw converter handles that aliasing detail within the 92 pixel Nyquist diameter. To keep that piece of additional information (as proof that the lens is limiting the resolution, not the print), I generally draw a 92 pixel diameter circle, but when it distracts, indeed one can cover the center with a disk.

What Arthur's shots also show is how contrast reduces as the detail approached its resolution limit, just like we can see in an MTF curve. With proper deconvolution sharpening the micro detail contrast will be increased and the resolution that is there will be easier to see, without exaggerating things. The resulting images will look more crisp, more realistic (because now our eyes will become the limiting factor, just like in real life).

Cheers,
Bart

[Bart_van_der_Wolf]

I did the test to make sure I want to keep this system. I might as well post it for others to see. It may be my imagination, or the weak low pass filter on the sonys but diffraction seems to jump in fast on the APS-C sonys.

Hi Arthur,

It just means that the system is accurate enough to show that diffraction will progressively reduce contrast towards the Nyquist limit, which is exactly what happens. At the wide open side of the aperture range contrast may (additionally) be limited by residual lens aberrations that are overpowering the diffraction effects, but diffraction is always present.

When the input detail has lower contrast than this average target contrast of almost 100:1, then it will hit the zero contrast output level (which is beyond recovery) faster.

Cheers,
Bart

[xpatUSA]

. . . also gives information about whether the print is offering reasonably more detail than the lens can resolve, and how the Raw converter handles that aliasing detail within the 92 pixel Nyquist diameter. To keep that piece of additional information (as proof that the lens is limiting the resolution, not the print), I generally draw a 92 pixel diameter circle, but when it distracts, indeed one can cover the center with a disk.

Thanks Bart,
Probably a regular question for you:
Is the physical diameter of that circle constant or does it depend on sensor spatial res. and taking distance? I understand that taking distance is not important within reasonable limits and why that is so but I'm thinking that the diameter of the 'Nyquist circle' in actual mm would vary?

[edit] I guess it all depends on what is meant by 'pixel', original target image (gif) pixels or the captured image (sensor) pixels. Now I'm getting muddled up . . any artifacts inside the said Nyquist circle are sensor (or printer) related are to do with the camera and not a lens under test. Which is why I find lens testing on my 9.12um sensor pixel pitch somewhat less than revealing . . . lots of moir[ay] though!

With proper deconvolution sharpening the micro detail contrast will be increased and the resolution that is there will be easier to see, without exaggerating things. The resulting images will look more crisp, more realistic (because now our eyes will become the limiting factor, just like in real life).

Interesting comment about our eyes, with which I agree. Makes you wonder how this gentleman, talking about Sigma Merrill shots, is viewing his images . . .

http://forums.dpreview.com/forums/thread/3450887

Well, he does mention "looking at an image at 100%" so that might be a clue  

Ted

[bjanes]

The target itself is already a challenge for the printer (and can be used for that as well, using a loupe and calipers to measure the blur diameter), but that also gives information about whether the print is offering reasonably more detail than the lens can resolve, and how the Raw converter handles that aliasing detail within the 92 pixel Nyquist diameter. To keep that piece of additional information (as proof that the lens is limiting the resolution, not the print), I generally draw a 92 pixel diameter circle, but when it distracts, indeed one can cover the center with a disk.

Bart,

Since your target is a challenge for the printer, wouldn't it make sense to provide a target at higher resolution so that one could print at a size larger than 130x130 mm when using the maximum resolution of the printer? For example, I am using the Epson 3880 which has a native resolution of 720 ppi, so if I could print at 260x260 mm at this resolution, I would effectively have twice the resolution. My printer does resolve the 3 resolutions at all contrast levels on your revised chart, but the lines are not that sharply defined.

Regards,

Bill

[Bart_van_der_Wolf]

Thanks Bart,
Probably a regular question for you:
Is the physical diameter of that circle constant or does it depend on sensor spatial res. and taking distance? I understand that taking distance is not important within reasonable limits and why that is so but I'm thinking that the diameter of the 'Nyquist circle' in actual mm would vary?

[edit] I guess it depends on what is meant by 'pixels', original target image pixels or the captured image pixels.

Hi Ted,

The diameter of the observed/captured blur disc is set by the limiting resolution of the observer/capture device. When shot with a discrete sampling device such as an image sensor, it will be constant, a minimum of 91.67 pixels in diameter at the Nyquist frequency for a target with 144 cycles (144 / Pi x 2). The sensor resolution is limited at it's Nyquist frequency, no smaller detail then 2 pixels per cycle (= 0.5 cycles/pixel) can be reliably recorded, but detail is often reduced to zero contrast at slightly larger detail size. There may be some accidental alignment of finer detail with a single sensel, causing a local high or low, but that will produce aliasing (unless the contrast is reduced, e.g. by an AA-filter) detail will seem to grow in size which will look as a deviation in predictable patterns.

So it's the captured image pixels that will show a constant diameter limit. The printed target will exhibit (sinusoidal grating + a few 1/2/3 linewidth) features all the way down to the pixel limits of the printer (where it struggles and reveals the effects of diffusion and loss of color accuracy due to lack of sufficient ink colors for dithering).

As long as these real features are shot with a small enough magnification factor due to shooting distance (>25x focal length), those smallest features will be smaller than the Nyquist frequency can resolve. Unfortunately, the artifacts at the printing limits will be relatively easy to spot, because they became larger than the finest details, a bit similar like aliasing is larger than the finest detail, and darker due to ink diffusion. Ink diffusion will make shadow detail blend together, and the target will predict how serious it can be.

Cheers,
Bart

[Bart_van_der_Wolf]

Since your target is a challenge for the printer, wouldn't it make sense to provide a target at higher resolution so that one could print at a size larger than 130x130 mm when using the maximum resolution of the printer? For example, I am using the Epson 3880 which has a native resolution of 720 ppi, so if I could print at 260x260 mm at this resolution, I would effectively have twice the resolution. My printer does resolve the 3 resolutions at all contrast levels on your revised chart, but the lines are not that sharply defined.

Hi Bill,

No, the printer is doing its best at 720 PPI, and is (barely) resolving detail and correct gray levels at that level (lack of ink colors to dither intermediate tones plus ink diffusion filling the paper white areas). That's not a problem, it's just struggling with physics. More accurate droplet placement, good head alignment, accurate paper transport, and good dithering can reduce this effect to a minimum, but its inherent to the process.

To make life easier for the printer, one could print the full target at 360 PPI effective resolution to a size of 260mm square, but then the finest detail that is smaller than 360PPI will also become larger, and the target will need to be shot at twice the distance to make sure that the lens/sensor cannot resolve this larger detail. That may cause impractical indoor shooting distances for longer focal lengths.

The Nyquist frequency of the captured shot will remain at 92 pixels diameter, regardless of the shooting distance (assuming finer detail than can be resolved due to magnification factor).

Cheers,
Bart

[Bart_van_der_Wolf]

Interesting comment about our eyes, with which I agree. Makes you wonder how this gentleman, talking about Sigma Merrill shots, is viewing his images . . .

http://forums.dpreview.com/forums/thread/3450887

Well, what he is experiencing is that aliasing at the limiting resolution of a discrete sampling device (especially when viewed on a low resolution display) looks a bit different than what we normally see with our eyes. What we can achieve with deconvolution, is restoration of real detail (assuming a more or less anti-aliased image).

Cheers,
Bart

[Fine_Art]

Here is the same f16 file as above with RL sharpening, micro-contrast on, contrast by detail on. If I really needed to mess with it I would export to IPlus. I think the fine detail is as much as you can ask for like this. More would start to look unnatural.

[Fine_Art]

Hi Arthur,

It just means that the system is accurate enough to show that diffraction will progressively reduce contrast towards the Nyquist limit, which is exactly what happens. At the wide open side of the aperture range contrast may (additionally) be limited by residual lens aberrations that are overpowering the diffraction effects, but diffraction is always present.

When the input detail has lower contrast than this average target contrast of almost 100:1, then it will hit the zero contrast output level (which is beyond recovery) faster.

Cheers,
Bart

Correct me if my understanding is off here. The diffraction spot size is a function of aperture so using a given size like f8 will show the effects faster on the smaller pixels of an APS-C sensor than on an FF like this D600.
D600 5.95 micron pixels
A55   4.78 micron pixels

Airy disk at f8 is 10.7 microns so it spills over into adjacent pixels on both cameras. F16 is 21.5 microns. Diffraction should be an issue.

Looking at the unsharpened images they have a similar detail extinction radius. They will all sharpen to about the same amount.

[Fine_Art]

Of course I didn't go out to just do the resolution charts.    Here is a real shot with a 100% crop from the far left side. f16 with focus on the front tree.

[Bart_van_der_Wolf]

Correct me if my understanding is off here. The diffraction spot size is a function of aperture so using a given size like f8 will show the effects faster on the smaller pixels of an APS-C sensor than on an FF like this D600.
D600 5.95 micron pixels
A55   4.78 micron pixels

Airy disk at f8 is 10.7 microns so it spills over into adjacent pixels on both cameras. F16 is 21.5 microns. Diffraction should be an issue.

That's correct. The absolute size of the diffraction pattern (usually measured at the diameter of the first zero of the Airy disk pattern) has a given size, determined by the aperture number and wavelength. Sensor arrays with a smaller sensel pitch will therefore use more sensels to sample that same diffraction pattern. That also means that the center weighted sampling of the blur pattern will be more accurate (which is good for deconvolution), but since more samples get averaged per unit area the individual sensels will be blurrier.

Looking at the unsharpened images they have a similar detail extinction radius. They will all sharpen to about the same amount.

That is not exactly what I see. I see lower contrast signals near Nyquist, and a slightly larger bur diameter as the aperture gets narrower (although I don't understand the shot labeled as f/16). On your f/1.8 shot I'd call it a 93x93 pixel blur diameter, at f/8 it's a 94 or 95 pixel diameter, and at f/16 I'd give it a 91x93 pixel diameter (which is frankly a bit strange if it really is an f/16 shot, there is way too much aliasing for an f/16 aperture). RawTherapee is good, but it can't defy physics.

Since it is not the full resolution pattern that I proposed, but only a square star area, I cannot verify if it was printed at a high enough resolution, or shot at a large enough distance. So maybe the features are larger than they should be and less challenging for the sensor, making the sensor set the limit and not the target?

Cheers,
Bart

[Fine_Art]

I did use your pattern with the edge blocks cropped off. This let me print it bigger on a 600dpi laser 8.5"x11" sheet. I up-rezed it in PS to generate the required number of pixels. In the top left corner of the screenshot you can see the full image where the piece of paper is a small rectangle. Anyone can calculate range with the focal length from that. I would estimate from memory 70 to 80 ft away. I am far enough away that the whole center is blurred out. I am sure that is not an issue. I can send the raws to you or someone at the site for people to convert with their own preferred software. The shots are of no artistic merit so copyright is not an issue! I can also convert a raw image you send me to see if it is the software.

I get similar close to nyquist with my Sony macros or the G lens. the kit zoom or other zooms are mush in comparison. I believe all sonys have weak aa filters.

This is the reason i made the post as A - diffraction should be an issue, B - the pictures show not much difference, A and B should not both be correct unless diffraction is not the overriding issue.

In my past study of artifacts (the color noise thing) I went down the rathole of looking at a wide assortment of de-bayer methods. It quickly became a mess of academic papers that were beyond where I wanted to go. One useful thing did come out of that, an idea of the difficulties in de-bayering. I suspect the local area search of the best algorithms followed by rules of interpretation may be making the diffraction issue secondary. Especially when lines are involved predictions (fake detail) are pushed forward. This will tend to beat the standard ISO resolution type charts.

Here is a link to the version of RT I am using, feel free to try it.
http://www.rawtherapee.com/forum/viewtopic.php?f=15&t=4507