DPReviewTV: What is diffraction in photography?

[dilbert]

But that was simply referring to life in general, now you limit that statement to “ So, light bending inside your camera is something you will never see.”, of course this again is false. What effect is there on light when it goes from one medium to another? Say, air to a piece of glass, or two pieces of glass with different refractory properties? The kind of stuff you’d find inside a lens..... It bends.

If "camera" is used here to refer to "camera body" then it is technically correct (except if the rear of the lens has protruded back into the cavity.) Otherwise, yes, you're right. Light bends, and different colours bend by different amounts. Hence rainbows.

 

[gruhl28]

Several problems with this demonstration. First and foremost, light does NOT bend. And discussion of light that justifies some light property by saying that light "bends" is automatically incorrect. Period. Second, using liquid wave tables or sound waves to describe what the action of electromagnetic radiation is like is not really representative.

If light doesn't bend, then what do lenses do, what is refraction? What would you say diffraction is caused by if not light bending? How would you explain interference patterns if light doesn't bend? For many situations, electromagnetic waves do behave analogously to liquid waves and sound waves.

 

[Joules]

If "camera" is used here to refer to "camera body" then it is technically correct (except if the rear of the lens has protruded back into the cavity.)

That would be cherry picking the meaning though. Keep in mind the poster previously said this:

There is nowhere on earth that anyone or any instrumentation will ever see light bend.

Which makes it quite clear that there must be some misunderstanding when it comes to the use of the word 'bend'. Unless the poster also temporarily forgot about all the devices we use day to day to bend light.

 

[usern4cr]

I don’t know if this helps but I did post a thread on diffraction.

Diffraction, Airy Disks and implications

Two points of light are clearly resolved when they are separated by distances that are much larger than the radius of the disk. As the separation decreases, the disks start overlapping and the resolution decreases. When the separation is the same as the radius, the disks have coalesced. Closer...

www.canonrumors.com

Thanks for the post, AlanF, but after reading the thread I didn't see any of the comments that would get into what the pupil size is for a long telephoto lens. If there was a link to a formula that would show this then I missed it. As the focal length becomes extremely long such as on a 10" wide 2500mm f10 telescope (without an adjustable iris/aperture blades), it must have a very large pupil diameter, on the order of an inch or so. I can't imagine that light going through a 1" wide hole can have any appreciable diffraction at all since the vast majority of the light is nowhere near the iris edge. That leads me to believe that the f# that causes appreciable diffraction should increase with the focal length of the lens. That's what I want to know as it would make all the difference when people talk about diffraction limits on long lenses.

 

[privatebydesign]

If "camera" is used here to refer to "camera body" then it is technically correct (except if the rear of the lens has protruded back into the cavity.) Otherwise, yes, you're right. Light bends, and different colours bend by different amounts. Hence rainbows.

Actually even technically it isn't correct as all sensors in cameras are stacks with some form of glass on the top of them, yes the light is traveling in a pretty collimated path by then but the edges have to bend more. Hmm I wonder if that is why performance of all lenses drops off the further from center you go.....

Almost like we are discovering stuff that has been know for years!

And you can never have to many rainbows

 

[privatebydesign]

Thanks for the post, AlanF, but after reading the thread I didn't see any of the comments that would get into what the pupil size is for a long telephoto lens. If there was a link to a formula that would show this then I missed it. As the focal length becomes extremely long such as on a 10" wide 2500mm f10 telescope (without an adjustable iris/aperture blades), it must have a very large pupil diameter, on the order of an inch or so. I can't imagine that light going through a 1" wide hole can have any appreciable diffraction at all since the vast majority of the light is nowhere near the iris edge. That leads me to believe that the f# that causes appreciable diffraction should increase with the focal length of the lens. That's what I want to know as it would make all the difference when people talk about diffraction limits on long lenses.

I don't understand what you are calling a "pupil diameter", the diameter of the apparent aperture is the focal length in mm divided by the f number. Your 2,500mm f10 telescope has an apparent aperture opening 2500/10 = 250mm diameter.

But my understanding is diffraction is a function of f value (and magnification) not apparent aperture size. Well obviously it is more complicated than that, it is a function of airy disc size in relation to magnification, that is why diffraction changes with the same lens on a different sized sensor.

 

[usern4cr]

The ‘fixed pupil’, apparent aperture opening, is 800/11 in mm, rounded out that means it is 73mm or 2.87 inches.

Looking down a lens and trying to estimate anything is an exercise in futility as the lenses you are looking through change the very view of what it is you think you are seeing.

As for how light is affected by such a ‘large hole’, well that is just the nature of waves and edges. But as I pointed out previously, the example shown was as extreme as it is possible to get even in a specialized situation and really can’t be replicated in non macro real world situations. I think few would argue the example at f2.8 was not sharp yet that equates to a non macro real world f17, more closed down than you 800 f11!

The "entrance pupil" is 800/11 = 73mm. The entrance pupil is the diameter of the central unobstructed light bundle coming in without considering any bending of the lens. But the lens bends/focuses the light so that it is much smaller as it reaches the iris/aperture blades. So the diameter of the pupil at the iris (which I called the "pupil" or "fixed pupil") is much smaller than the "entrance pupil". It is the diameter of this "pupil" that I'm interested in.

I agree that looking down the lens can be misleading, but I wouldn't call it futile since it is to be expected to act that way.

What I am trying to get at is how f11 on a 15mm lens will act vs on a 100mm lens or on a 2500mm lens, focused at infinity for simplicity of comparison. I would expect it to behave drastically differently as far as the amount of noticeable diffraction it causes. I noticed on your additional post that diffraction is a function of f# and magnification. That's getting into what I'm talking about.

 

[Ian K]

No it won't. I have two really big issues with the video, first, he is demonstrating a visual diffraction between f17 and f96, at which point f17 looks pretty darn good. Besides, I don't know any Canon lenses that stop down past f32.

Pretty much all of the canon lenses that can take a doubler will become f/64 with the 2x extender attached.

 

[Joules]

What I am trying to get at is how f11 on a 15mm lens will act vs on a 100mm lens or on a 2500mm lens, focused at infinity for simplicity of comparison. I would expect it to behave drastically differently as far as the amount of noticeable diffraction it causes. I noticed on your additional post that diffraction is a function of f# and magnification. That's getting into what I'm talking about.

My understanding is that f-number (as in, 24-70 mm f/4.0) is all you need to be concerned with when you want to judge how apparent diffraction will be at a given digital magnification. The last part there is just saying that it gets more apparent as you zoom into your picture.

But the magnification that is actually affecting the Airy disk size is already baked into the f-number, as that combines physical aperture and focal length.

The other magnification that can make things confusing is the distance to your subject, but you were talking about focus at infinity. At which, based on my understanding, you should expect no difference in the amount of diffraction blur you see regardless of focal length. Only f-number matters.

At close distances, the aperture on the lens is not the effective aperture anymore, which means distance comes into play as well. That was part of the video in the OP.

 

[privatebydesign]

Pretty much all of the canon lenses that can take a doubler will become f/64 with the 2x extender attached.

Good point, though I wonder how many people are buying tele's, especially fast and expensive ones, and then using them at f32?

 

[Nemorino]

I don't know any Canon lenses that stop down past f32

Afaik the 100-400 stops down to f/38 at the long end and two TS Makro even down to f/45 (90mm and 135mm).

 

[usern4cr]

My understanding is that f-number (as in, 24-70 mm f/4.0) is all you need to be concerned with when you want to judge how apparent diffraction will be at a given digital magnification. The last part there is just saying that it gets more apparent as you zoom into your picture.

But the magnification that is actually affecting the Airy disk size is already baked into the f-number, as that combines physical aperture and focal length.

The other magnification that can make things confusing is the distance to your subject, but you were talking about focus at infinity. At which, based on my understanding, you should expect no difference in the amount of diffraction blur you see regardless of focal length. Only f-number matters.

At close distances, the aperture on the lens is not the effective aperture anymore, which means distance comes into play as well. That was part of the video in the OP.

Wow - After looking it up online, it does seem that you're right - it's just so surprising to me. I guess the future R5s might not see much better photos than the R5 at high f#'s for any focal length lenses after all. I guess that might give more "value" to the high IQ wide open lenses like the 85mm f1.2 than they currently have for the R5 since you might really need exceptionally high IQ & low f# lenses to really see a big benefit for the doubly bigger R5s MP's. I wonder if it'll come with quad pixel AF, or if they'll only have that (initially) in their R1 (at lower MP's) as a big selling point for the R1?

 

[Joules]

Wow - After looking it up online, it does seem that you're right - it's just so surprising to me. I guess the future R5s will not see much better photos than the R5 at high f#'s for any focal length lenses after all. I guess that might give more "value" to the high IQ wide open lenses like the 85mm f1.2 than they currently have for the R5 since you might really need exceptionally high IQ & low f# lenses to really see a big benefit. I wonder if it'll come with quad pixel AF, or if they'll only have that (initially) in their R1 (at lower MP's) as a big selling point for the R1?

The really wide apertures like f/2.0 and wider are typically not used in conditions that you would associate with a desire for maximum detail though. So I don't know if the 85 mm 1.2 is such a great example.

But I don't feel like the R5 s (high res R) is all that threatened yet. It is essentially just the same pixel density as the 90D, if it actually is ~ 90 MP. The R5 begins to show softening at f/9.0 and a 90 MP FF body at 6.3 according to the Photo pills calculator. That's still an absolutely common aperture for wildlife, where all that extra detail and room for cropping will be very much appreciated.

And in landscape photography, if you are concerned about maximizing detail, you should focus stack anyway, making it okay to shoot with a wider aperture.

Also worth noting is that you can actually reduce the blur quite a bit using deconvolution techniques - for example with Photoshop smart sharpen oder Canon's DLO. Yes, it can't restore detail actually lost due to overlapping Airy disks. But it can make details more visible that were no longer visually identifiable.

Another thing to consider is that it would actually be ideal to be diffraction limited all the time. As it is right now, AA (or low pass) filters are essentially necessary to introduce some blur artificially. And there's false detail in our images anyway due to having sensors that are not capable of fully resolving what the lenses do. This is again mostly relevant for wildlife, especially birds. Where fine detail and patterns in the feathers tens to get messed up with aliasing and artifacts.

The other way to think about high resolution is like this: For each expensive lens you buy, as long as you are not using a body that is diffraction limited with this lens, you are not getting the most value for your money. Not a big deal if detail isn't all that important, but for those expensive big whites, I think it is something positive to have in the back of your head about diffraction.

 

[AlanF]

The really wide apertures like f/2.0 and wider are typically not used in conditions that you would associate with a desire for maximum detail though. So I don't know if the 85 mm 1.2 is such a great example.

But I don't feel like the R5 s (high res R) is all that threatened yet. It is essentially just the same pixel density as the 90D, if it actually is ~ 90 MP. The R5 begins to show softening at f/9.0 and a 90 MP FF body at 6.3 according to the Photo pills calculator. That's still an absolutely common aperture for wildlife, where all that extra detail and room for cropping will be very much appreciated.

And in landscape photography, if you are concerned about maximizing detail, you should focus stack anyway, making it okay to shoot with a wider aperture.

Also worth noting is that you can actually reduce the blur quite a bit using deconvolution techniques - for example with Photoshop smart sharpen oder Canon's DLO. Yes, it can't restore detail actually lost due to overlapping Airy disks. But it can make details more visible that were no longer visually identifiable.

Another thing to consider is that it would actually be ideal to be diffraction limited all the time. As it is right now, AA (or low pass) filters are essentially necessary to introduce some blur artificially. And there's false detail in our images anyway due to having sensors that are not capable of fully resolving what the lenses do. This is again mostly relevant for wildlife, especially birds. Where fine detail and patterns in the feathers tens to get messed up with aliasing and artifacts.

The other way to think about high resolution is like this: For each expensive lens you buy, as long as you are not using a body that is diffraction limited with this lens, you are not getting the most value for your money. Not a big deal if detail isn't all that important, but for those expensive big whites, I think it is something positive to have in the back of your head about diffraction.

In theory you can restore detail lost to overlapping Airy discs if you know the point spread function. The higher the pixel density, the better my 400 f/4 performs for resolution relative to the 100-400 f/5.6 or 100-500mm f/7.1. As the resolution of the sensor gets higher and higher, what determines resolution ends up being the diameter of the entrance pupil of the lens and not the focal length or f number.

 

[Joules]

In theory you can restore detail lost to overlapping Airy discs if you know the point spread function.

But in practice, we don't know it and there are other, more practical techniques of imaging past the diffraction limit from what I understand. Though the ones I'm thinking of are also not practical for everyday photography.

what determines resolution ends up being the diameter of the entrance pupil of the lens and not the focal length or f number.

What's the difference between these? Isn't f-number just the result of focal length/entrance pupil?

Edit: Do you mean resolution in the sense of the smallest physical detail you can make out from a given distance, for example details on the moon? Then of course you are right, physical aperture is the key factor there.

 

[usern4cr]

But in practice, we don't know it and there are other, more practical techniques of imaging past the diffraction limit from what I understand. Though the ones I'm thinking of are also not practical for everyday photography.


What's the difference between these? Isn't f-number just the result of focal length/entrance pupil?

Edit: Do you mean resolution in the sense of the smallest physical detail you can make out from a given distance, for example details on the moon? Then of course you are right, physical aperture is the key factor there.

From what I've read the entrance_pupil size limits the possible angular detail possible (eg for astronomy), but it is the f# that limits the airy disc size and not the entrance_pupil alone.

As far as getting the most resolution past the airy disc size, I don't use Adobe products so I can't take advantage of the techniques you mention, but the DXO PL4 prime output does give me some remarkable output for almost no effort - but I don't know how it compares with what you can do.

 

[dilbert]

Actually even technically it isn't correct as all sensors in cameras are stacks with some form of glass on the top of them, yes the light is traveling in a pretty collimated path by then but the edges have to bend more. Hmm I wonder if that is why performance of all lenses drops off the further from center you go.....

Ah, I forgot about that. Yes, the sensor directly under lens would be better than sensor at the edges.

 

[AlanF]

But in practice, we don't know it and there are other, more practical techniques of imaging past the diffraction limit from what I understand. Though the ones I'm thinking of are also not practical for everyday photography.


What's the difference between these? Isn't f-number just the result of focal length/entrance pupil?

Edit: Do you mean resolution in the sense of the smallest physical detail you can make out from a given distance, for example details on the moon? Then of course you are right, physical aperture is the key factor there.

Yes, i mean the the smallest detail that can be resolved as in your edit.

 

[AlanF]

From what I've read the entrance_pupil size limits the possible angular detail possible (eg for astronomy), but it is the f# that limits the airy disc size and not the entrance_pupil alone.

As far as getting the most resolution past the airy disc size, I don't use Adobe products so I can't take advantage of the techniques you mention, but the DXO PL4 prime output does give me some remarkable output for almost no effort - but I don't know how it compares with what you can do.

I find that DxO PL4 gives me better resolution than the Adobe products.

 

[usern4cr]

I find that DxO PL4 gives me better resolution than the Adobe products.

Really? Wow - that's sooo great to hear, since I'm a fellow PL4 user.

I've always been completely amazed at how the "Prime" output could take my grainy (but sharp) Olympus M43 photos and output them with super smooth beautiful backgrounds with great detail. Now I can do the same with my even better R5 photos, and not feel like I'm missing out since I avoid the Adobe products.