Lens diffraction - truth or urban legend?

Stewart, for the comparison to be meaningful you should repeat it with the same ISO setting and adjust only F-Stop and shutter speed. Otherwise we'll never know how the distance scale vanished.

And- I really like my Two EP2's.
 
Brian Sweeney said:
Stewart, for the comparison to be meaningful you should repeat it with the same ISO setting and adjust only F-Stop and shutter speed. Otherwise we'll never know how the distance scale vanished.

And- I really like my Two EP2's.
Click to expand...

Ah .... I'll need a teenager to do that, she's out at the moment.


the first photo is the f22 version btw
 
I figured the first was the F22 version, and is also the one shot at ISO 400. Rather than declare "Look diffraction erased the distance scale", let's try to get the gain on the sensor out of the equation.

And the best way to determine the differences is to register the two images and do a frame-to-frame subtract.
 
Just for giggles, another 4/3 to APS-C comparison:

Leica-Panasonic 45/2.8 macro on G1 vs. Nikon 50/1.8D on D300.

Spoiler: at f/22, both lenses deliver ~750 lp/ph (MTF-50) right out to the corners — just like the Olympus 50/2 and Canon 100/2.8 macros shown in the similar comparison above.

Conclusion: at apertures of f/11 and smaller, where diffraction is undeniably limiting, APS-C (1.4 or 1.5 crop) and 4/3 (or micro 4/3) sensors deliver effectively identical resolution and microcontrast, from center to corners. At wider apertures, where lens aberrations rather than diffraction are limiting, the results obtained depend on the specific lens used, and on the sensor density. The best results obtained with both systems are, again, nearly indistinguishable.

This is an even smaller difference than the already small difference predicted by theory. I don't know why this is the case, but one can compare many different lens/sensor combinations over at DP Review, and these general conclusions are remarkably robust.

With squarish aspect ratios (4:3, 5:4, square) there is in practice effectively no difference at all between APS-C and 4/3. With wider aspect ratios the APS-C systems should have a slight advantage (roughly 12.5% for 3:2 vs. 4:3) across the long edge of the frame.

It will be interesting to see how well these conclusions hold up at wider apertures (say, f/4), as pixels continue to shrink on cameras like the 18 megapixel Canon 7D and the projected 15 megapixel 4/3 and micro 4/3 cameras that many of us expect will be released this year.
 
Last edited:
I just see it

I just see it

I don't go intentionally looking for it, and I don't see it in all u4/3 images, or even most. But when I do, and someone is asking for feedback, I have mentioned it a couple of times.

I'm sorry that you don't or can't see it, or recognize it. Maybe that's a good thing for you?


semilog said:
We've now established that (1) you can do arithmetic; and (2) you get the same answers as me and everyone else who can also do arithmetic. Congratulations.

Here is the question that you are avoiding: how do you get from there to being "able to see diffraction" in micro 4/3 pictures at web resolution (I'll be generous: 800 pixels wide), without (you claim) pixel peeping?

(Sneak preview: you don't.)

Here's the problem that you face: At f/22, with a 10 mpix 4/3 sensor, the Olympus Zuiko 50 mm macro delivers a measured resolution (MTF-50) corresponding to >740 line pairs/ph along the picture's short axis (and hence 1000 lp/ph along the long axis).

That's at f/22, as far as the lens stops down.

Here are the measurements.


In other words, no, you can't see diffraction in images from a 4/3 sensor at web resoution - unless by "web resolution" you mean images larger than 1500 x 2000 pixels, and images taken at f/22 or smaller.

At a more reasonable aperture, f/8, the same lens is Nyquist-limited (right out to the corners!) for the same 3648 x 2736 sensor.

So you can "see" the diffraction limitation of images (and separate this limitation from other sources of noise and contrast degradation) taken on 4/3 cameras on web-based images, exactly how?

I'm deeply skeptical.
Click to expand...
 
interesting link

interesting link

thanks for providing it.

On my monitor, it is clear that both systems are diffraction limited at (the default aperture at your link below) though the 4/3 looks worse on my monitor.

Here's where you can actually see the mtf lp/mm drop for the 4/3 by 30-40% vs the APS - change just the aperture on each to f4 or f5.6. The APS retains its resolution by a huge margin, while the 4/3 drops off like a cliff.


semilog said:
Just for giggles, another 4/3 to APS-C comparison:

Leica-Panasonic 45/2.8 macro on G1 vs. Nikon 50/1.8D on D300.

Spoiler: at f/22, both lenses deliver ~750 lp/ph (MTF-50) right out to the corners — just like the Olympus 50/2 and Canon 100/2.8 macros shown in the similar comparison above.

Conclusion: at apertures of f/11 and smaller, where diffraction is undeniably limiting, APS-C (1.4 or 1.5 crop) and 4/3 (or micro 4/3) sensors deliver effectively identical resolution and microcontrast, from center to corners. At wider apertures, where lens aberrations rather than diffraction are limiting, the results obtained depend on the specific lens used, and on the sensor density. The best results obtained with both systems are, again, nearly indistinguishable.

This is an even smaller difference than the already small difference predicted by theory. I don't know why this is the case, but one can compare many different lens/sensor combinations over at DP Review, and these general conclusions are remarkably robust.

With squarish aspect ratios (4:3, 5:4, square) there is in practice effectively no difference at all between APS-C and 4/3. With wider aspect ratios the APS-C systems should have a slight advantage (roughly 12.5% for 3:2 vs. 4:3) across the long edge of the frame.

It will be interesting to see how well these conclusions hold up at wider apertures (say, f/4), as pixels continue to shrink on cameras like the 18 megapixel Canon 7D and the projected 15 megapixel 4/3 and micro 4/3 cameras that many of us expect will be released this year.
Click to expand...
 
Stewart

Stewart

Maybe toss in some bright greens, reds, yellows - will be very easy to see.

Sparrow said:
Well you know my skill with digi-cams, it was on auto it did one at f22 1/30 and 400 asa, the other f5.6 1/40 at 100 asa, for whatever reason

The lighting was the same on both
Click to expand...
 
Hi Stewart

Hi Stewart

There's a stop or two difference here, I am guessing that maybe the G1 has a max shutter exposure and it timed out? I think you'll want to do the tests outdoors or in brighter light so the f22 one doesn't timeout.

also, putting some greens, reds, and yellows in the photo will make it even easier to pick out.

Thanks.

Sparrow said:
Yep me too, an example would be good, I set up the worst combination I have, my 12mm and my daughters GF1, the camera was on auto and while I can see there are differences between f5.6 and f22 I'm unsure what I'm actually looking at

this was the set-up

...
Click to expand...
 
OK, both at 100 asa, ignore the motion blur Alice interpreted "keep still" as "burst out laughing"

f22




f5.6



click the pic for the original image, what do you see? I don't understand the registration thing Brian, sorry
 
I see it

I see it

the reduced contrast and resolution on your daughters face and hair (despite the motion blur) can be readily seen.

But to see it even easier, expose with a bit more light, and put some red, green, and yellow items in the frame.

Sparrow said:
OK, both at 100 asa, ignore the motion blur Alice interpreted "keep still" as "burst out laughing"


click the pic for the original image, what do you see? I don't understand the registration thing Brian, sorry
Click to expand...
 
In the 1980s I did a lot of image processing, "machine vision". Registering two images usually meant finding features in each, and using those features to "register" or line the images up. One image typically had to be shifted and be resampled to line-up within sub-pixel resolution. After that was done, just subtract one from the other and you get a new image that is the difference frame.

Cute pictures of your daughter, nice socks. Nikki has not hit the "sock phase" yet, but is big into multi-colored finger-nails.

Looking at Stewart's first set of shots, I wonder how many pictures taken stopped down also cause the ISO to be set higher by the camera's auto exposure system. This would certainly make the shots look noisier than those taken wide-open.
 
Last edited:
ampguy said:
the reduced contrast and resolution on your daughters face and hair (despite the motion blur) can be readily seen.

But to see it even easier, expose with a bit more light, and put some red, green, and yellow items in the frame.
Click to expand...

You can readily see it in the blurred area?... sorry, I think you are talking rubbish
 
Brian Sweeney said:
In the 1980s I did a lot of image processing, "machine vision". Registering two images usually meant finding features in each, and using those features to "register" or line the images up. One image typically had to be shifted and be resampled to line-up within sub-pixel resolution. After that was done, just subtract one from the other and you get a new image that is the difference frame.

Cute pictures of your daughter, nice socks. Nikki has not hit the "sock phase" yet, but is big into multi-colored finger-nails.

Looking at Stewart's first set of shots, I wonder how many pictures taken stopped down also cause the ISO to be set higher by the camera's auto exposure system. This would certainly make the shots look noisier than those taken wide-open.
Click to expand...

I have an old application that will do that subtraction, but I'm not sure it will work with files this size, Ill look tomorrow

On simply a subjective inspection the f22 version of that one seems to have the better resolution but slightly reduced contrast to my eye, what do you think?
 
On the two photographs shown, one at F22 and the other at F5.6, rather than trying to fool anyone that I have super-vision and can readily tell one from another on the basis on contrast and resolution:

Stewart, your daughter is beautiful and these are very cute pictures.
 
since we got onto digital sensors, can someone tell me if my thinking on this is correct. Taking n M9 which has 5212 pixels in the horizontal dimension. Then dividing by 36 (the mm width) we get 144. Therefore in the horizontal dimension we can only get 72lp/mm which is less than half that achievable with B+W film on 35mm when using a tripod. Is this correct ?
 
You also have to factor in the Bayer-pattern Mosaic filter on the M9, and the color of the lines. It can be as few as 36lp/mm. Other sensors have anti-aliasing filters which reduces the resolution to less than just the raw pixel count.

I have one camera that is monochrome and calculating resolution on it is 2pixels per line pair.
 
ampguy said:
If you have a really good lens, line prs per mm is theoretically best wide open:

you can measure it, and you can see it. If you are seeing more line prs per mm stopped down, and are accurately measuring, consider upgrading lenses.


f/ line pairs per mm

1.4 1,100...
Click to expand...

Oh wow. Lemme borrow your diffraction-limited 35/1.4... 😀
 
tlitody said:
since we got onto digital sensors, can someone tell me if my thinking on this is correct. Taking n M9 which has 5212 pixels in the horizontal dimension. Then dividing by 36 (the mm width) we get 144. Therefore in the horizontal dimension we can only get 72lp/mm which is less than half that achievable with B+W film on 35mm when using a tripod. Is this correct ?
Click to expand...

Right, and it can be visible if you wet print from film and compare it to digital printing from M9... But if instead of wet printing you scan film and digitally print it, M9's printed files can be as good as scanned film printed files because there's quality loss when we scan... For real great tone and resolving power, wet print medium format...

Cheers,

Juan
 
You must log in or register to reply here.
Back
Top Bottom