Canon 50D and Macro

Hello and apologies at the start for bringing up a subject that has been discussed often before.

I have used a 20D with Canon 100mm macro for some time for macro-photography, mostly of insects, and a D60 and 10D before that. I've had the intent of adding a 40D, but now that the 50D with its 15 megapixel sensor has arrived, I'm not sure what to do. I'd like to get the 50D, but diffraction scares me. As I understand the situation, the 20D becomes "diffraction limited" (the term I've seen used to indicate that it starts to become visible) at about f11. I use f11 to f16 routinely without apparent problem. At above f16 I can detect perhaps a little softening, but only when I view images at 100%, so I usually don't do it.

I had anticipated the slightly worse "diffraction limit" at about f10 for the 40D, but from what I can figure out there will be a much bigger hit for the 50D at about f8. Extrapolating from my experience with 20D, then, would I expect that over f11 diffraction would be visible on the 50D for the same type of subjects? My guess is that the answer is yes, but I don't understand the theory nearly well enough to be certain of that.

Thanks.

I've seen this sort of confusion a lot, and it stems from a misperception of what "diffraction limitation" means. I've seen many comments to the effect that "Camera X with its small pixels is diffraction limited at f11, but camera Y with its bigger pixels is only diffraction limited at f16". The implication is that camera Y is "less limited" by diffraction, and therefore somehow is producing sharper images less affected by diffraction.

This is incorrect. Loosely speaking, diffraction produces a spreading of light rays due to the finite size of the lens aperture; the amount of spreading is the same for a given aperture no matter what size pixels are in the sensor sitting on the focal plane of the camera -- it's a property of the lens optics, not the sensor.

What the sensor does is to record the image including the effects of the diffraction; the diffraction produces a slight blurring of fine detail. As usual, more pixels mean more resolution, and even though slightly blurred, the image is still resolved better with more pixels.

The mistake typically made is to think of the 100% pixel-level view is somehow absolute, and used to judge resolution (and noise) in images. But the monitor has fixed size pixels, and so 100% view is accomplished by magnifying the smaller pixels of camera X more than the larger pixels of camera Y. So 100% view is not absolute, it's relative to the size of the pixels in the sensor.

Quantitative details: To make a correct statement about resolution, one should use absolute measures of resolution such as how closely spaced two lines can be (say in mm) and still be resolved by the sensor. So absolute resolution measures are given in lines or line pairs/mm.

DPReview tests lenses on both the 40D and 5D, two different format sensors with two different size pixels as well; they have tested the 70-200/f2.8:
http://www.dpreview.com/lensreviews/wid ... on.xml%3F3
The 5D has 12MP, and a hypothetical FF sensor with pixels the size of the 40D would have over 25MP; let's call this hypothetical camera the 1Ds Mk40D. Since we imagine two sensors of the same FF size, the 5D and the 1Ds Mk40D, the resolution in lp/mm and line pairs/picture height are equivalent. While I will discuss two full frame sensors, the issue is pixel size relative to frame size, and so analogous statements hold for the 40D and 50D with their differing pixel counts for the same sensor size.

What we have in hand are resolution tests on the APS-C 40D and the FF 5D, in line pairs/picture height, and the picture height in mm is different for the 40D and 5D so they are not comparable measures. So we make them comparable by dividing each by the frame height in mm (resolution of N lp/ph times 1 ph/24mm gives resolution in lp/mm for the 5D, and resolution of M lp/ph times 1 ph/14.8mm for the 40D gives resolution in lp/mm for the 1Ds Mk40D).

One other potential pitfall -- the DPR tests give resolution at different points across the frame, as a percentage of frame size from the center to the corners. But away from the center by a given percentage is imaging a different part of the lens on the 40D than it is on the 5D; one must either compensate for that, or more simply just use the center results since that's the same for both.

So using the center resolution values in lp/ph for the DPR test of the 70-200/2.8, I divided by the respective ph's in mm and found that the 1Ds Mk40D has higher resolution than the 5D at ALL apertures, even when the optics is severely diffraction limited. The main effect of diffraction is to limit the amount of improvement to be got from smaller pixels; the improvement is over 40% at f5.6, but only about 10% at f16 and beyond. But it's always an improvement.

Bottom line: Smaller pixels provide higher resolution at all apertures, even in the "diffraction limited" regime. The emphasis on this regime is misleading, and results in the common misperception of diffraction limitation as a problem. I would turn it around and say that the key issue is instead "pixel pitch limitation" -- larger pixel size results in a wider range of apertures where the limiting factor in resolution is the pixel size rather than the optics. Smaller pixel pitch reduces the range of apertures where the limiting factor in resolution is the sensor rather than the optics, and that is a good thing.

emil
Just to clarify my thinking about what diffraction is (or what it is caused by).
To use a parallel idea - sunlight when it strikes the edge of an object that has thickness casts a shadow. The edge of the shadow is not a sharp crisp line of dark to light but has a gray transition of dark to light at the edge of the shadow. This is caused by the sunlight "bending" slightly around the edge of the object, which in turn causes this gray edge of the shadow.

So if I understand diffraction correctly - as you close down the aperture this "bending" of light around the edge of a small aperture become a greater portion of the light that is transmitted to the sensor. This "bent" light is not in the same critical focus as the rest of the light coming through the lens so it tends to softens the image. The smaller the aperture the great the amount of "bent light vs. focused light.

In lay terms is this correct?

Now for a question - Why is diffraction more of a problem with sensors than with film?
Is it because film is for all intents a flat surface while a sensor is not flat, it has depth?

signgrap wrote:emil
Just to clarify my thinking about what diffraction is (or what it is caused by).
To use a parallel idea - sunlight when it strikes the edge of an object that has thickness casts a shadow. The edge of the shadow is not a sharp crisp line of dark to light but has a gray transition of dark to light at the edge of the shadow. This is caused by the sunlight "bending" slightly around the edge of the object, which in turn causes this gray edge of the shadow.

So if I understand diffraction correctly - as you close down the aperture this "bending" of light around the edge of a small aperture become a greater portion of the light that is transmitted to the sensor. This "bent" light is not in the same critical focus as the rest of the light coming through the lens so it tends to softens the image. The smaller the aperture the great the amount of "bent light vs. focused light.

In lay terms is this correct?

Now for a question - Why is diffraction more of a problem with sensors than with film?
Is it because film is for all intents a flat surface while a sensor is not flat, it has depth?

Dick - That thinking is pretty much correct, except you get slightly different effects with light passing through the small aperture, rather than bending around a corner of an object casting a shadow. Check out this Cambridge in Colour short tutorial on diffraction. The diagrams at the top are a great help I think. Especially showing how diffraction (as you noted) always happen with light and corners or edges, regardless of aperture.

I don't have any experience with film, but I think I know the answer. And that's that diffraction certainly does occur with film, and can affect it. Film has a sort-of random grain structure however, that likely may disguise the diffraction limits to some extent. Plus, 35 mm film is like having large pixel full frame digital, in that the diffraction limit is up in the f/16-f/22 range (I think), a smaller aperture than most people usually use. If you had some super fine grained, high resolution, film, you would also see diffraction effects more easily, just like with smaller pixel sensors.

Thanks for taking the time for such a detailed reply. I believe that it is finally sinking in. I have a followup question, however, which is: if, as you say, smaller pixels provide higher resolution at all apertures, why is diffraction raised as a problem so often when a camera gets a large increase in number of pixels, as with the 50D? It is quite striking the number of posts on various forums today that complain that 15 megapixels is too many!

Fred

Emil, to me this is about the easiest to understand explanation of the diffraction issue I've come across.... thanks!

Romy

ejmartin wrote:I've seen this sort of confusion a lot, and it stems from a misperception of what "diffraction limitation" means. I've seen many comments to the effect that "Camera X with its small pixels is diffraction limited at f11, but camera Y with its bigger pixels is only diffraction limited at f16". The implication is that camera Y is "less limited" by diffraction, and therefore somehow is producing sharper images less affected by diffraction.

This is incorrect. Loosely speaking, diffraction produces a spreading of light rays due to the finite size of the lens aperture; the amount of spreading is the same for a given aperture no matter what size pixels are in the sensor sitting on the focal plane of the camera -- it's a property of the lens optics, not the sensor.

What the sensor does is to record the image including the effects of the diffraction; the diffraction produces a slight blurring of fine detail. As usual, more pixels mean more resolution, and even though slightly blurred, the image is still resolved better with more pixels.

The mistake typically made is to think of the 100% pixel-level view is somehow absolute, and used to judge resolution (and noise) in images. But the monitor has fixed size pixels, and so 100% view is accomplished by magnifying the smaller pixels of camera X more than the larger pixels of camera Y. So 100% view is not absolute, it's relative to the size of the pixels in the sensor.

Quantitative details: To make a correct statement about resolution, one should use absolute measures of resolution such as how closely spaced two lines can be (say in mm) and still be resolved by the sensor. So absolute resolution measures are given in lines or line pairs/mm.

DPReview tests lenses on both the 40D and 5D, two different format sensors with two different size pixels as well; they have tested the 70-200/f2.8:
http://www.dpreview.com/lensreviews/wid ... on.xml%3F3
The 5D has 12MP, and a hypothetical FF sensor with pixels the size of the 40D would have over 25MP; let's call this hypothetical camera the 1Ds Mk40D. Since we imagine two sensors of the same FF size, the 5D and the 1Ds Mk40D, the resolution in lp/mm and line pairs/picture height are equivalent. While I will discuss two full frame sensors, the issue is pixel size relative to frame size, and so analogous statements hold for the 40D and 50D with their differing pixel counts for the same sensor size.

What we have in hand are resolution tests on the APS-C 40D and the FF 5D, in line pairs/picture height, and the picture height in mm is different for the 40D and 5D so they are not comparable measures. So we make them comparable by dividing each by the frame height in mm (resolution of N lp/ph times 1 ph/24mm gives resolution in lp/mm for the 5D, and resolution of M lp/ph times 1 ph/14.8mm for the 40D gives resolution in lp/mm for the 1Ds Mk40D).

One other potential pitfall -- the DPR tests give resolution at different points across the frame, as a percentage of frame size from the center to the corners. But away from the center by a given percentage is imaging a different part of the lens on the 40D than it is on the 5D; one must either compensate for that, or more simply just use the center results since that's the same for both.

So using the center resolution values in lp/ph for the DPR test of the 70-200/2.8, I divided by the respective ph's in mm and found that the 1Ds Mk40D has higher resolution than the 5D at ALL apertures, even when the optics is severely diffraction limited. The main effect of diffraction is to limit the amount of improvement to be got from smaller pixels; the improvement is over 40% at f5.6, but only about 10% at f16 and beyond. But it's always an improvement.

Bottom line: Smaller pixels provide higher resolution at all apertures, even in the "diffraction limited" regime. The emphasis on this regime is misleading, and results in the common misperception of diffraction limitation as a problem. I would turn it around and say that the key issue is instead "pixel pitch limitation" -- larger pixel size results in a wider range of apertures where the limiting factor in resolution is the pixel size rather than the optics. Smaller pixel pitch reduces the range of apertures where the limiting factor in resolution is the sensor rather than the optics, and that is a good thing.

signgrap wrote:emil
Just to clarify my thinking about what diffraction is (or what it is caused by).
To use a parallel idea - sunlight when it strikes the edge of an object that has thickness casts a shadow. The edge of the shadow is not a sharp crisp line of dark to light but has a gray transition of dark to light at the edge of the shadow. This is caused by the sunlight "bending" slightly around the edge of the object, which in turn causes this gray edge of the shadow.

So if I understand diffraction correctly - as you close down the aperture this "bending" of light around the edge of a small aperture become a greater portion of the light that is transmitted to the sensor. This "bent" light is not in the same critical focus as the rest of the light coming through the lens so it tends to softens the image. The smaller the aperture the great the amount of "bent light vs. focused light.

In lay terms is this correct?

Yes, in that diffraction is a spreading of the direction of light propagation due to the finite aperture it passes through. The diffraction that blurs a sharp shadow into a finite transition region is indeed also a diffraction effect. Where I would diverge from your explanation is the statement that it comes from light passing near the edge of the aperture. Diffraction is a wave phenomenon, resulting from the interference of wave components from all across the aperture, not just at the edges. This is true both for diffraction from an aperture and from the edge of an object casting a shadow.

Any wave obeys the uncertainty principle -- that its position and direction of motion cannot both be determined with infinite precision, and that the more one tries to define one aspect, the more uncertain is the other. Presenting light with a finite aperture to pass through localizes its position in the plane of the aperture, more and more as the aperture is closed down; consequently its direction of motion after passing through the aperture becomes more and more uncertain, and this is the spreading of the light beam that increases as the aperture decreases that one observes with higher f-stop.

Now for a question - Why is diffraction more of a problem with sensors than with film?
Is it because film is for all intents a flat surface while a sensor is not flat, it has depth?

I don't buy the premise of the question. Again, diffraction is a property of the optics, not the medium used to record the image. Finer grained film would be able to exhibit the effects of diffraction at wider apertures than coarser grained film, by being better able to resolve the diffraction spot; yet AFAIK nobody used to obsess about fine grained film being "more diffraction limited" than coarse grain film. And why the obsession with diffraction? Why not say that smaller pixels or film grain are more "chromatic aberration limited", or more "coma limited", at wide apertures, since they will exhibit these effects as well, beginning at larger f-number as one goes toward the wide end? The point is that these are all properties of the optics in front of the sensor; all that a finer grained film or smaller pixel sensor does, is record the optical effects of lens aberrations and diffraction more faithfully.

Very nice explanation Emil. Thanks.

emil, David

Thanks for you responses they have helped a lot.
So diffraction is a lens design issue - As the f-stop decreases in size, the diffracted (non-focused) light passing through the lens becomes a larger portion of the light striking the sensor, causing a softening of the image. At some point (different for each lens design) the gains made in DOF by reducing the f-stop are offset by the loses caused by diffraction softening.

emil I under now that the computer and 100% pixel views have really made this a discussion about sensors as in the days of film you didn't normally have the ability to view grain @ 1:1. Digital has made it so cheap (after your up front costs) to take pictures that doing tests for diffraction is easy.

Thanks

Dick - It's not really so different for each lens design. f/4 on a 50 mm lens will have the same size aperture no matter what design or manufacturer it is; And the diffraction effect will be the same. (There may be very minor differences in diffraction effect do to the shape of the aperture opening - they aren't perfect circles - but I doubt you'd be able to tell in the image.)

The way I usually think about 100% views is to think of what equivalent print size it would be. If I think something looks bad from a five megapixel image at 100%, I remind myself that that's the equivalent of a 30" print (or something like that, depending on the size of your monitor and number of pixels), so of course it looks noisy/grainy/blurry/etc.

I'm not sure this will help but perhaps you'd like a look at the issue in the telescope world? I'm not really sure that "diffraction limit" as a term of art is used the same way in photography as in telescope evaluation, but FWIW:

http://www.astro.cornell.edu/academics/ ... _limit.htm

http://cse.ssl.berkeley.edu/bmendez/ay1 ... /lec8.html

Cheers jim