Fairy tales about 'diffraction'

[agorabasta]

The fact is that diffraction is wavelength-dependent. The fact is that the blurring at tight apertures is not wavelength-dependent as the said blurring shows no colour fringes.

So, quite unequivocally, the tight-aperture blur does not come from diffraction.

It actually comes from a much more mundane process - scattering. That scattering happens on all optical imperfections/non-uniformities in the light path. The greatest of such imperfections are the surfaces of the lens elements. So the more elements, the more scattering happens, and the faster is the image degradation at aperture stop-down; and we all know that from experience with consumer-grade superzoom lenses.
The blurring mechanism is very simple. When the lens is wide open, the light rays inside it are not localised, every part of the lens elements transmits some part of the light to hit a particular point on the sensor; it means that the scattering is causing an evenly spread haze that slightly decreases the contrast. Then if the aperture is tightened, the rays to hit a particular point/sensel become localised already as they traverse the lens elements; hence the scattered light is also localised around the ray of light off which that light is scattered. You then see the loss of local contrast and acuity that is so universally and erroneously ascribed to diffraction.

The about only kind of lenses showing the real effect of diffraction are the high-quality primes. Personally, I have only one lens that really exhibits diffraction on stop-down, it's a Sigma EX 30/1.4. And that diffraction first shows as increased colour noise around sharp edges and/or in the sharply detailed textures, and it also shows as erratic jagginess of the fine luma detail. These effects actually increase the subjective sharpness of the image, much like the perceived sharpness is increased with the added grain. Then it takes stopping the lens too far to just get the diffraction limiting the acuity, but it never actually happens as the scattering effect is stronger anyway.

But sure there are some special optical devices/schemes that really hit the diffraction limit. It's just not the case with off-the-shelf lenses and large-sensor cameras.

[agorabasta]

If that sounded like some overly techy blah-blah, here's a very simple explanation.

At tight apertures there's only a very narrow path leading from a point in the scene to a point on the sensor. So all the imperfections on that path distort the image formed on the sensor.
At wide apertures there are many paths of the same narrow width adding together, so parts of registered light take different paths and take different distortions. Then the registered light is a sum of light from all the paths, while the distortions, being random, do partially cancel each other out.

So essentially, the lens's glass (glass, plastics, coatings) is more transparent and clear at wide apertures.

[twm47099]

I ran some quick tests with my Minolta 100 f/2.8 and my A700. I started at f/2.8 and shot down to f/32. The subject was a filmy white curtain at about 10 ft. I still have to collect all the shots and up load them, and I will probably want to repeat using a tripod. The shots I took were focused manually and I was sitting so I didn't change focus between shots. I have only looked at them under magnification on the rear LCD.

But with those limitations, it appears that the fine lines in the curtain become sharper at about f/8 and at f/16 they are definitely disappearing. At f/32 there is a mush. Their appeared to be rainbow colors at the lines as they softened, but I will have to examine them at higher mag to see if there is any banding.

Since the plot of diffraction intensity vs distance from center of a point is a curve with a large peak followed by decreasing peaks (exponential ??) as distance increases for each wavelength, I don't expect there will be too many pure colors.

tom

[agorabasta]

Since the plot of diffraction intensity vs distance from center of a point is a curve with a large peak followed by decreasing peaks (exponential ??) as distance increases for each wavelength, I don't expect there will be too many pure colors.

You don't need the secondary maximum, you just need the first minimum of diffraction pattern of specific colour to cover a sensel of that colour, and you immediately get a strongly coloured spot. There's the Bayer pattern, remember?

[Dr. Harout]

Agorabasta, your second post, the more "scientific" one is very interesting. As for the first, it was so "easy" that...

[agorabasta]

He-he... Here's one more bit. At tight apertures it's not only the light paths in the lens that get too narrow, it's also the light paths in the air between the lens and the remote object that also get too narrow. So the air impurities and temperature/density variations also add to those scattering distortions. Hence the images of the most remote objects lose sharpness faster than the images of the closer objects. And you may readily notice that the far objects expected to fall well within the DOF actually get blurred out too fast with aperture tightening.

[01af]

Sorry I didn't read this forum for a while so I am late to this discussion ... but I don't buy the 'scattering' theory.

That said, agorabasta isn't entirely wrong. It's true that diffraction is wavelength-dependent, so all other things being equal, red light will get diffracted more than blue light. The amount of diffraction is proportional to the wavelength. It also is true that, in principle, different amounts of diffraction for different colours should result in colour fringes around high-contrast edges. Finally, a phenomenon like scattering really exists and will be more pronounced at smaller apertures indeed.

However agorabasta's conclusions from these facts are wrong. First, the differences in the amounts of diffraction for the various colours simply are way too small to show up as actual colour fringes in a digital image. Second, the scattering of light is wavelength-dependent, too—that's why, among other things, the sky is blue.

To see how much water the colour fringing argument is holding, just load one of your own images into Photoshop (or into whatever image-processing software you're using). Pick an image taken with a good lens (prime lens or pro-grade zoom) at or near the lens' optimal aperture, i. e. between f/4 and f/8, that contains a lot of fine detail like e. g. distant buildings or power-line pylons. Prefer a raw conversion over an in-camera JPEG file in order to avoid even the tiniest JPEG compression artefacts which may distort this little experiment. If applicable, remove any colour fringes coming from chromatic aberration as neatly as you can. Now go into the RGB channels palette, select the red channel, and apply Gaussian blur to it with a radius of, say 1.6 pixels. Do the same to the green channel but with a radius of 1.4 pixels. Finally, blur the blue channel with a radius of 1.1 pixels. This will mimic diffraction blur that you'd get when shooting at an aperture of approx. f/32 or f/45 or thereabouts; it also simulates the different degrees of blur in proportion to the colours' wavelengths. Compare to the original image at full view, 100 % view, and, say, 400 % view.

When pixel-peeping then you'll notice that the added blur is not always perfectly colour-neutral indeed. However the colours will vary, depending on the original colours of the respective detail and its background, and you won't see any obvious colour fringes. Instead, what you see is mostly just an overall loss of micro-contrast and detail resolution—just as if there was diffraction at work.

Of course, you can provoke colour fringes through different amounts of blur applied to the individual RGB channels if you try hard enough ... but to accomplish that, you'd need much more variation in the pixel radius of the Gaussian Blur filter. Try this, for example: Blue 1 pixel; Green 2 pixels; Red 6 pixels. Now you have it: nasty colour fringes everywhere. But please note how far the strengths of the blur are out of proportion now with regard to the respective wavelengths. Also note how the fringes still are not too obvious when looking at the whole picture (as opposed to pixel-peeping).

So, the bottom line is this: The absence of obvious colour fringes in 'small-aperture blur' does not prove this kind of blur cannot be caused by diffraction. And scattering of light through imperfections on the lens elements' surfaces or in the air inside the lens and camera is a few orders of magnitude smaller, so it cannot be used to explain 'small-aperture blur'.

[Birma]

Very interesting explanation and experiment Olaf. Nice to see you back on the forum

[Dr. Harout]

Seconded

[agorabasta]

Modelling diffraction blur with a gaussian blur is a brain-dead idea.

Diffraction creates secondary peaks AND troughs off the central spot. So at the diffraction threshold you get certain higher spacial frequencies suppressed while the lower spacial frequencies are amplified. E.g. if the secondary peak from the spot source covers the adjacent pixels of the green diagonal array, you get blur in the green channel; then if the secondary peak moves one pixel farther, you get a primary trough covering the immediate adjacent green pixels thus boosting the highest spacial freq's. So the diffraction is both suppressing and boosting sharpness near the threshold, essentially redistributing the signal energy between higher freq components of the image.

A more sane 'experiment' would be to move the sharpening radius slider in ACR while keeping the other parameters fixed; you'd see then the subjective overall sharpness increasing with an increased blur radius, although the finest detail is obliterated. Exactly that is done by diffraction as the primary trough radius increases.

So - no, you don't get diffraction effects with 99% of the lenses available; they simply are not transparent enough. The blur you get with stopping down is of same nature as the blur/haze you get, say, from some grease/dirt on the lens front element. The glass is simply not good enough to get you the visible diffraction effects.

[01af]

Modelling diffraction blur with a gaussian blur is a brain-dead idea.

No, it's not. Of course I agree it's not a very accurate model—but good enough to see a point or two ...

Diffraction creates secondary peaks AND troughs off the central spot.

The problem is, you are wasting way too much thoughts on the details of the shape of a single Airy disk without fully understanding what it is and what it does and what not. So you fail to see the forest for the trees. When Airy disks are where points are supposed to be then this will basically result in blur. As simple as that.

For example if the secondary peak from the spot source covers the adjacent pixels of the green diagonal array, you get blur in the green channel; then if the secondary peak moves one pixel farther, you get a primary trough covering the immediate adjacent green pixels thus boosting the highest spacial frequencies.

Yeah—but did it ever occur to you that digital images are made up by more than one Airy disk? You see ... the Airy disk is the tree, and the blurred image is the forest. There are Airy disks everywhere—not just where green pixels are ... not even just where any pixels are. Think about it!

So the diffraction is both suppressing and boosting sharpness near the threshold ...

Diffraction never boosts sharpness; it always only suppresses it. And what is this 'threshold' you're fantasizing about? There are no thresholds anywhere. Diffraction is a continuous phenomenon.

A more sane 'experiment' would be to move the sharpening radius slider in ACR while keeping the other parameters fixed; you'd see then the subjective overall sharpness increasing with an increased blur radius, although the finest detail is obliterated. Exactly that is done by diffraction as the primary trough radius increases.

Sorry, that's just bunkum. The radius parameter in the unsharp-mask algorithm does an entirely different thing that has nothing to do with diffraction.

I just took two pictures with the Minolta AF Macro 50 mm 1:3.5 on the Sony A900, one at f/5.6, the other at f/32, from a tripod at (almost) infinity distance, in order to compare real 'small-aperture blur' to my artificially blurred models thereof. I found two things.

First, a Gaussian blur with radii of Red 1.6, Green1.4, Blue 1.1 pixels does not resemble the diffraction blur of f/32 (as I suggested yesterday); it rather is similar to the diffraction blur at f/45 - f/64 or thereabouts. To mimic diffraction blur at f/32, Gaussian blur with a radius of 1 pixel on a full-size 24 MP image out of a 35-mm-format camera is about right.

Second, real diffraction in fact is not entirely colour-neutral—so this pulls the carpet from under agorabasta's feet. Dark detail against a bright background does have a very faint colour shift towards red. So this puts an end to this bizarre discussion. After all, if agorabasta actually was right then text books on physics and optics would have to be rewritten, and all those physicists, professors, and engineers of optics designing and building lenses, telescopes, and astronomer's observatories were just fools failing to understand what diffraction really is.

By the way, adding Gaussian blur with radii of Red 1.1 - 1.2, Green 1.0, Blue 0.8 - 0.9 pixels to the picture taken at f/5.6 simulates both the blur and the faint colour shift of the diffracted picture taken at f/32 fairly well. The simulation isn't perfect, of course, because the modelling of the diffraction through Gaussian blur is rather coarse ... but it does point into the right direction. In particular, in the simulation the red shift is more local while in the diffracted image it's more distributed. Moreover, the f/5.6 image is cleaner while in the f/32 image there's more luminance noise. This might be a hint that agorabasta's 'scattering' also is at work ... but just as well it might be just a difference in the sensor noises at different shutter speeds (1/160 s vs 1/5 s at ISO 160/23°). Remember—longer shutter speeds tend to be noisier. Anyway, I never said 'scattering' didn't exist; I'm just saying it's not the reason for 'small-aperture blur'. Diffraction is.

[agorabasta]

After all, if agorabasta actually was right then text books on physics and optics would have to be rewritten, and all those physicists, professors, and engineers of optics designing and building lenses, telescopes, and astronomer's observatories were just fools failing to understand what diffraction really is.

The optical tools most readily showing diffraction effects are the optical microscopes, like those you should have seen at elementary school. Seems like you have stopped learning well before that because otherwise you'd clearly know what diffraction looks like. And then there are those fine coloured patterns in the reflections off all fine-grain surfaces you obviously never had enough curiosity to look into.

Another thing you so stubbornly refuse to recognise is that the digital camera sensors normally employ Bayer filters, and those differently sized diffraction patterns in the individual colour channels get strongly aliased with also differently distributed individual colour arrays. Hence the diffraction along the image sharp edges gets a lot of noise-like false colour and luma space aliasing. And there's a clear threshold for this to happen as the diameter of the primary trough gets slightly above the sensel diagonal.

And I will be back to tear down the rest of your delusions as there are too many of them, about everything you've written above in the thread is dead wrong and is also presented in an impolite manner.

[01af]

After all, if agorabasta actually was right then text books on physics and optics would have to be rewritten, and all those physicists, professors, and engineers of optics designing and building lenses, telescopes, and astronomer's observatories were just fools failing to understand what diffraction really is.

The optical tools most readily showing diffraction effects are the optical microscopes ...

Oh yes, sure, and microscopes.

Do you really believe those who are making microscopes don't understand what they are dealing with? There are theories and formulas precisely describing the reasons and the effects of diffraction. Just read a basic text book on physics. And these formulas don't hold for microscopes only but also for telescopes and lenses. Among other things, these formulas predict the degree of blur to be expected in photographs taken at small apertures—et voilà, there it is in a picture taken with a 35-mm-format camera at f/32. Just open your eyes and see. There's a reason why there is a strong correlation between the image format and the smallest apertures of most lenses intended for that format. For 35-mm format, that's f/16 - f/22. Beyond these apertures, diffraction blur will start to cross the boundary from 'just barely visible' to 'objectionable'. Macro lenses offer somewhat smaller apertures still because in the context of extremely narrow depth-of-field at high magnifications, the objectionability of diffration blur will shift a bit.

Scattering also is well understood among physicists. So there also are formulas to describe the interactions between small particles (e. g. molecules) and light. The physicsts can precisely calculate which colour the sky would be when Earth's atmosphere wasn't mostly nitrogen and oxygen but, say, methane and ammonia.

And then comes Imperial Ambassador agorabasta and tells us they all just don't understand what really is going on and he's the only one in the know. Hey, come on—time to wake up and get a life!

And then there are those fine-coloured patterns in the reflections off all fine-grain surfaces you obviously never had enough curiosity to look into.

Now you're confusing interference and diffraction.

Another thing you so stubbornly refuse to recognise is that the digital camera sensors normally employ Bayer filters, and those differently sized diffraction patterns in the individual colour channels get strongly aliased with also differently distributed individual colour arrays. Hence the diffraction along the image sharp edges gets a lot of noise-like false colour and luma space aliasing.

Digital imaging sensors not only have Bayer colour filter arrays; they also have anti-aliasing filters. So don't worry about the aliasing of Airy disks' structures with individual photosites. It mostly doesn't happen. And if it did then the resulting colour artefacts would not show up as colour fringes but as random patterns of colours.

Obviously you never looked at a picture suffering from diffraction blur. Diffraction in a stopped-down photo lens never emphasises any spatial frequencies; that's just in your fantasy. It always blurs spatial frequencies—higher frequencies more, lower frequencies less. So diffraction doesn't act in an aliasing way ... to the contrary, it acts in the same way as the anti-aliasing filter does. In a diffraction-blurred image there's always less moiré, not more.

And there's a clear threshold for this to happen as the diameter of the primary trough gets slightly above the sensel diagonal.

Here, you are falling for the usual misconception about the interaction of two resolution-limited channels where the output of the first is the input to the second. Actually, diffraction will visibly compromise an image's sharpness long before the Airy disk's primary trough's diameter comes anywhere close to the pixel pitch.

And I will be back to tear down the rest of your delusions as there are too many of them ...

You're welcome. But maybe you should attend to your own delusions first, as they are way more numerous than mine.

[agorabasta]

And then there are those fine-coloured patterns in the reflections off all fine-grain surfaces you obviously never had enough curiosity to look into.

Now you're confusing interference and diffraction.

It's exactly you who is supposed to know that diffraction is nothing more than interference

[agorabasta]

And if it did then the resulting colour artefacts would not show up as colour fringes but as random patterns of colours.

And it shows exactly like that, just like I stated above. And I also mentioned above that the only lens I have observed it with was a Sigma 30/1.4, and it was on a a700 body with a strong AA filter.