[Gooshin]

Does it ever occur to you when watching your pics that "would be nice to have a little more in the frame (wider view)"

please, read the whole thread.

[Tokina]

please, read the whole thread.

What's the point? Feel free to explain what is wrong in my thinking.

[Gooshin]

nothing is wrong with your thinking

its just that what you are thinking has been mentioned, questioned, discussed, confirmed and noted 3-4 times already in this thread, and if you continue with your line of thinking we are going to start going in circles again.

I am eagerly awaiting what Falconeye will say as he and Wheatfield have steered this never ending discussion into a more interesting direction.

[falconeye]

What were your testing methods and test parameters (TOC)?

I am eagerly awaiting what Falconeye will say

I feel urged now to respond Thanks for the interest.
Wheatfield asks a very good question (how measure a lens' resolution beyond the sensor's resolution?) and here is my method:

Assume, you want to measure a lens A: Reverse-mount lens A (with a filter-to-filter-screw adapter) to another lens B with significantly longer focal (say the difference in focal length is a factor M like M=6x) and mount the long lens B to the body as usual. Focus both lenses at infinity. Place a subject with tiny details in front of lens A, at a distance of the focal length of A. Note that, because all lenses are focussed at infinity, both lenses are operated at their sweet spot for resolution. This is not challenging either lens' macro capabilities!

Because everything captured by lens A is magnified by M, the resolution figures for lens B can be worse by a factor M and the quality of lens B doesn't matter as long as if it fully exploits the sensor's capabilities.

Assume A=Zeiss 50/1.4, B=DA*300/4, M=6, K20D (5µm pixel pitch). Then, theoretically, a detail just 0.8µm large should fill one pixel. *IF* lens A has a resolution of 0.8µm. If not, you'll see the real resolution capabilities of lens A in its focal plane (and near the center). This is so because by some fundamental law of optics, light rays are always reversible in direction.

Now, what is the outcome? Well, see at the attached image yourself. It is a 100% crop (so, each pixel corresponds to 0.8µm in the real world). The image was taken at f/4.0. The green lines are 1.5µm wide and 3µm apart and are created by interference of (green) light. Red light interference leads to a somewhat coarser patterns, as expected. I see no blue light pattern, so it either may not have been resolved or it may not have existed in the subject. A 3µm wide line pair corresponds to 333 lp/mm and the contrast still is very high! This is more or less the diffraction limit at f/4 (which is at 1.29 µm)! Zeiss quotes 300-400 lp/mm for their 35mm primes and I confirm it. In comparison, the K20D sensor requires 100 lp/mm.

BTW, what you see is a part of a bee's eye and you see its many segments reflecting some sun light (the segments are invisible to the naked eye...). The photo is mine.

BTW, 1.5µm are 2 * SQRT(2) * lambda_green and I think the interference is coming from double reflection of light at the gap where two segments meet. But I am not sure. It could be reflection from something within the eye segments as well...

BTW, if you would like to repeat my measurement: Please take into account that the combo of lens A and B behaves as a lens like 1800mm f/24. So, you basically have NO light and no chance to use long exposure! 1800mm shakes like hell even on a tripod. I had to use a tripod, sunlight, high ISO AND staking many frames together to come up with a result. And did I talk about focussing?

Check out this post from Yvon Bourque. I think your statement about all DA lenses not working on FF sensors is incorrect:[/url]

Good link. May I refer to our own information here, too?
https://www.pentaxforums.com/forums/pentax-slr-lens-discussion/31629-da-lens-...ts-thread.html

---

Last edited by falconeye; 06-15-2011 at 05:28 AM.

[Wheatfield]

I feel urged now to respond Thanks for the interest.
Wheatfield asks a very good question (how measure a lens' resolution beyond the sensor's resolution?) and here is my method:

<snip>

Thanks for the response. I did something similar a while back with my computer screen:



I don't think the magnification is as high as yours, but you do have the DSLR crop factor advantage. Mine was shot with an LX.

Anyway, it's pretty well known that a good lens will have very high aerial resolutions, in the range of 400 or more lppm, but in pictorial use, the resolution can drop to ~1/10 of that.

If you check Yoshihoko's lens resolution site, you will find the very best that the lenses tested were able to muster was under 100lppm using a USAF test target (theoretically a knife edge target). In pictorial use, resolution is generally 60% or so of a high contrast target.
Other testers are getting similar results, so his results are not one off.

Simple mathematics puts the K20 sensor at just over 66 lppm for 3 colours, or just under 200 lppm on an individual pixel basis.

Unless I am missing something (which is likely), it seems to me that the K20 is already out resolving pretty much any lens available to us.

I welcome your thoughts on this.

[Lowell Goudge]

Thanks for the response. I did something similar a while back with my computer screen:
Pug Gallery
I don't think the magnification is as high as yours, but you do have the DSLR crop factor advantage. Mine was shot with an LX.

Anyway, it's pretty well known that a good lens will have very high aerial resolutions, in the range of 400 or more lppm, but in pictorial use, the resolution can drop to ~1/10 of that.

If you check Yoshihoko's lens resolution site, you will find the very best that the lenses tested were able to muster was under 100lppm using a USAF test target (theoretically a knife edge target). In pictorial use, resolution is generally 60% or so of a high contrast target.
Other testers are getting similar results, so his results are not one off.

Simple mathematics puts the K20 sensor at just over 66 lppm for 3 colours, or just under 200 lppm on an individual pixel basis.

Unless I am missing something (which is likely), it seems to me that the K20 is already out resolving pretty much any lens available to us.

I welcome your thoughts on this.

I have stayed away from this thread for a long time, but, and I am not an expert on resolution by any means, if your assessment is correct, then the pixel pitch of the K20D sensor really defines the present state of the art of lenses, and implies that a realistic maximum image size full frame camera (if any one were to make it, should be 31.7 MP.

There are a coupe of things about this, first, no one is there yet! and second, for any one working on sensors, the real objective at this point should be targeted at improved dynamic range, better and higher ISO, and improved noise at all ISO, within the present pixel size, not increasing the pixel count, or spending the much higher production costs in producing a chip 2.25 times bigger in surface, considering the yeild, and limited use.

OK there will be wide angle fanatics who will want huge MP and full frame for landscapes, and interior/archetctural photography, but the present range of lenses for ASPC sensors really does suit 99% of the market now.

The other extreme would be to say, OK 15MP is sufficient, give us bigger photosites on the chip, but again, the larger surface area is a huge cost driver, not just because it is larger, but production yeild will drop because there is 2.25 times the possibility of a non recoverable flaw in the chip.

[jeffkrol]

Thanks for the response. I did something similar a while back with my computer screen:


I don't think the magnification is as high as yours, but you do have the DSLR crop factor advantage. Mine was shot with an LX.

Anyway, it's pretty well known that a good lens will have very high aerial resolutions, in the range of 400 or more lppm, but in pictorial use, the resolution can drop to ~1/10 of that.

If you check Yoshihoko's lens resolution site, you will find the very best that the lenses tested were able to muster was under 100lppm using a USAF test target (theoretically a knife edge target). In pictorial use, resolution is generally 60% or so of a high contrast target.
Other testers are getting similar results, so his results are not one off.

Simple mathematics puts the K20 sensor at just over 66 lppm for 3 colours, or just under 200 lppm on an individual pixel basis.

Unless I am missing something (which is likely), it seems to me that the K20 is already out resolving pretty much any lens available to us.

I welcome your thoughts on this.

Ahhh grey areas of the measurement world.......
In the real world, outside of industrial and scientific applications, you don't get lenses that are diffraction limited at f1.4. A more realisitc figure for a good quality photographic lens would be diffraction limited performance at f5.6 (or possibly f4.0) which would translate to 270 or 375 lines/mm respectively. A sensor that could fully resolve the resulting image would have 62MP or 120MP respectively which is clearly well beyond anything on the market today.
Re: When pixels meets the real world...: Open Talk Forum: Digital Photography Review
Or this cute one;
24MP is nice, and we're just getting there now. Yes, the 24MP FF might outresolve your WA zoom in the corners, wide open - whoopedy - doo dah ding, but it coarsely undersamples a super-tele or high-end macro at their sweet spots.
Re: When lens meets pixels...: Open Talk Forum: Digital Photography Review



Do Sensors &ldquo;Outresolve&rdquo; Lenses?

---

Last edited by jeffkrol; 11-12-2008 at 08:15 AM.

[reeftool]

Thanks for posting the link to the lens info, it looks like the DA300 would be a good buy if the world changes to FF but the 10-17 fisheye would add to my paperweight collection. My concern isn't if Pentax makes a FF camera which it will have to do if the market goes that way but if they would no longer support the current aps size sensor bodies. I don't have a K mount film camera, an SP1000 screw mount. While I have some M 42 lenses I don't use them often because my eyesight is getting a little off and auto focus lenses are a big help. Its only been a year since I finally made the jump to digital SLR and the rumors of a format change a a little disturbing as I am now making decisions on new glass I want to get.

[falconeye]

Thanks for the response. I did something similar a while back with my computer screen: [...] Anyway, it's pretty well known that a good lens will have very high aerial resolutions, in the range of 400 or more lppm, but in pictorial use, the resolution can drop to ~1/10 of that. [...] If you check Yoshihoko's lens resolution site [...] Simple mathematics puts the K20 sensor at just over 66 lppm for 3 colours, or just under 200 lppm on an individual pixel basis [...] it seems to me that the K20 is already out resolving pretty much any lens available to us.

First, I agree with subsequent comments from jeffkrol and Lowell Goudge.
I am also with Wheatfield as far as zoom lenses are concerned. But not for primes.
We already agree that good APS-C or FF primes resolve at 350 lp/mm which is the diffraction limit at f/4 (there is only one FF lens I am aware of which is diffraction limited at f/2.8 (Zeiss 35mm f/2)).
The Yoshihoko test, or any test site in this respect, publish resolution figures which are limited by the recording medium, be it film or sensor.

I have actually developped a mathematical framework to model published resolution figures from resolutions as limited by the medium, by diffraction, and by classical optics. You can solve those equations for classical optics performance and for the best primes, I empirically found a law saying that resolution r_oc as limited by classical optics is as follows:

r_oc ~ r0 * F

where r0 = 100 +/- 10 lp/mm and F be the f-stop number (in the range of about 1.4 to 11). BTW, there should be a constant and a square term but for some reason, the linear term dominates them.
Take into account the resolution r_oq limited by diffraction and r_m limited by the medium and you obtain the resulting resolution for any apertures and any sensor. Very telling... BTW, r_oq is known to be

r_oq = 1/2 / (1.22 * lamda/2 * F) = [1480 lp/mm] / F

Interestingly, the combined resolution r_oc,oq maxes out at

F = Sqrt([1480 lp/mm]/r0) or F=3.7 for r0 = 110 lp/mm

However, because I didn't yet publish my findings, I won't get into too much detail here. Just let's say that at f/4, a 350 lp/mm resolution is possible and I would estimate the Pentax FA 31mm to be cacapble of this. Without my mathematical framework (where AFAIK, there is no published equivalent) you won't be able to assess how a lens would perform with a sensor other than that used in the test. So, the Yoshihoko tests are meaningless in our context.

Having said this, your comment about "aerial resolution" vs. "resolution in pictorial use" triggers my curiosity. What do you mean by this and why should they differ? Are you talking about contrast? The contrast in both, your and my sample image, is great.

[Wheatfield]

First, I agree with subsequent comments from jeffkrol and Lowell Goudge.
I am also with Wheatfield as far as zoom lenses are concerned. But not for primes.
We already agree that good APS-C or FF primes resolve at 350 lp/mm which is the diffraction limit at f/4 (there is only one FF lens I am aware of which is diffraction limited at f/2.8 (Zeiss 35mm f/2)).
The Yoshihoko test, or any test site in this respect, publish resolution figures which are limited by the recording medium, be it film or sensor.

If you read Yoshi's test method, you will find he is using a standard 1000:1 target and T-Max 100 film.
At a TOC of 1000:1, Kodak publishes resolution numbers for that film as 200 lpm, my understanding is that Kodak treats the black lines in the chart as what is being resolved, and so they are claiming 200 lppm (a black line separated by a white space).
I don't have access to the ISO 6328 publication, so I can't confirm this directly.
This puts the measurement well within the T-Max 100 specification, but this is also a useless measurement for pictorial use, where a TOC of 1.6:1 is generally considered to be more useful, since we rarely photograph 10 stop edges in the real world.

I have actually developped a mathematical framework to model published resolution figures from resolutions as limited by the medium, by diffraction, and by classical optics. You can solve those equations for classical optics performance and for the best primes, I empirically found a law saying that resolution r_oc as limited by classical optics is as follows:

r_oc ~ r0 * F

where r0 = 100 +/- 10 lp/mm and F be the f-stop number (in the range of about 1.4 to 11). BTW, there should be a constant and a square term but for some reason, the linear term dominates them.
Take into account the resolution r_oq limited by diffraction and r_m limited by the medium and you obtain the resulting resolution for any apertures and any sensor. Very telling... BTW, r_oq is known to be

r_oq = 1/2 / (1.22 * lamda/2 * F) = [1480 lp/mm] / F

Interestingly, the combined resolution r_oc,oq maxes out at

F = Sqrt([1480 lp/mm]/r0) or F=3.7 for r0 = 110 lp/mm

However, because I didn't yet publish my findings, I won't get into too much detail here. Just let's say that at f/4, a 350 lp/mm resolution is possible and I would estimate the Pentax FA 31mm to be cacapble of this. Without my mathematical framework (where AFAIK, there is no published equivalent) you won't be able to assess how a lens would perform with a sensor other than that used in the test. So, the Yoshihoko tests are meaningless in our context.

I think my eyes are bleeding......
Having said that, as I am fairly certain that Yoshi's test is within the resolution limit of the test medium, I believe his numbers are relevant, even though we have switched capture media.
The lenses remain more or less the same.

Having said this, your comment about "aerial resolution" vs. "resolution in pictorial use" triggers my curiosity. What do you mean by this and why should they differ? Are you talking about contrast? The contrast in both, your and my sample image, is great.

My layman's understanding (I am not a physicist, and my termonology may be incorrect) is that you can get incredibly high resolutions off a lens if you ignore the fact that eventually it hits something (either film or sensor), and that real world resolution numbers are always a fraction of theoretical maximized resolution numbers.

Resolution charts used to be published with both 1000:1 and 1.6:1 figures, and the lower contrast numbers were always significantly lower (generally ~50%) of the high contrast numbers.
One question that I have never recieved a satisfactory answer to is: Are optical sensors (DSLR sensors) prone to the same resolution losses as film in lower contrast pictorial use?
My gut says no, but......

[stevebrot]

Wow! This thread has really gone stratospheric!

Now the more important question of all...how many line-pair can you fit on the head of a pin if photographed with a Pentax-FA 31/1.8 Limited?

Seriously though...if understand everyone properly...

Nah, there is NO WAY I CAN UNDERSTAND EVERYONE PROPERLY...

Steve

(...and I ain't dumb...a little slow, but not dumb...)

[stevebrot]

One question that I have never recieved a satisfactory answer to is: Are optical sensors (DSLR sensors) prone to the same resolution losses as film in lower contrast pictorial use?
My gut says no, but......

I see your line of thought here, but I guess a lot depends on the design of the sensor and the processing algorithms that are used handle the data. It is much the same type of issue as signal processing for digital music.

Steve

[stevebrot]

First, I agree with subsequent comments from jeffkrol and Lowell Goudge.
I am also with Wheatfield as far as zoom lenses are concerned. But not for primes.
We already agree that good APS-C or FF primes resolve at 350 lp/mm which is the diffraction limit at f/4 (there is only one FF lens I am aware of which is diffraction limited at f/2.8 (Zeiss 35mm f/2)).
The Yoshihoko test, or any test site in this respect, publish resolution figures which are limited by the recording medium, be it film or sensor.

I have actually developped a mathematical framework to model published resolution figures from resolutions as limited by the medium, by diffraction, and by classical optics. You can solve those equations for classical optics performance and for the best primes, I empirically found a law saying that resolution r_oc as limited by classical optics is as follows:

r_oc ~ r0 * F

where r0 = 100 +/- 10 lp/mm and F be the f-stop number (in the range of about 1.4 to 11). BTW, there should be a constant and a square term but for some reason, the linear term dominates them.
Take into account the resolution r_oq limited by diffraction and r_m limited by the medium and you obtain the resulting resolution for any apertures and any sensor. Very telling... BTW, r_oq is known to be

r_oq = 1/2 / (1.22 * lamda/2 * F) = [1480 lp/mm] / F

Interestingly, the combined resolution r_oc,oq maxes out at

F = Sqrt([1480 lp/mm]/r0) or F=3.7 for r0 = 110 lp/mm

However, because I didn't yet publish my findings, I won't get into too much detail here. Just let's say that at f/4, a 350 lp/mm resolution is possible and I would estimate the Pentax FA 31mm to be cacapble of this. Without my mathematical framework (where AFAIK, there is no published equivalent) you won't be able to assess how a lens would perform with a sensor other than that used in the test. So, the Yoshihoko tests are meaningless in our context.

Having said this, your comment about "aerial resolution" vs. "resolution in pictorial use" triggers my curiosity. What do you mean by this and why should they differ? Are you talking about contrast? The contrast in both, your and my sample image, is great.

Fascinating...when/where are you planning on publishing your study?

Steve

[falconeye]

If you read Yoshi's test method, you will find he is using a standard 1000:1 target and T-Max 100 film.

Well. T-Max 100 isn't even the best out there. T-Max 100 is specified to be 200 lp/mm (and you are right, it is line pairs as in the analog world, there was no distinction between lines per mm and lp/mm). In my post https://www.pentaxforums.com/forums/316914-post25.html I actually used an "Adox 20 CMS Pro" film which is specified at 800 lp/mm, or 4x the resolution of the Technical Pan which is on par or slightly better than the T-Max 100.

There have been studies trying to verify those claims (I don't remember the link). The outcome was that T-Max 100, Technical Pan etc. see their limit with test charts at ~100 lp/mm and so-called Gigabit films (like the Adox 20 CMS Pro) at ~150 lp/mm. This is still good as normal films resolve at 40-50 lp/mm. But the message is also simple: Don't trust the film guys' media resolution claims! They are plain wrong! 100 lp/mm is what a K20D delivers and *NO* film will outperform it except for dedicated "Gigabit films" which outperform it by a factor of 1.5x (only). I actually looked at the Adox 20 CMS test shots using a microscope and it does outperform the K20D. But only by a small margin. The true test of a lens is using my test method I described earlier (bee eye).

Go to Yoshihiko Takinami's test site, look up the best single test figure (which is110 lp/mm) and you know the true resolution of the T-Max film: 110 lp/mm. This is pretty much in line with other people's findings, including mine.

And if in doubt, just trust me

this is also a useless measurement for pictorial use, where a TOC of 1.6:1 is generally considered to be more useful, since we rarely photograph 10 stop edges in the real world.

Light is purely additive. So, in the absence of noise, the TOC (and I assume you mean the subject's contrast ratio by this) doesn't matter. However, there is a lot of noise (formerly known as grain ) with film and some noise with digital. The performance of DSLRs at ISO100 (FF even more than APS-C) however is excellent and the test chart resolution figures are meaningful because they are available even with low contrast subjects by using just a little bit of extra sharpening. But you are right, with increasing ISO, test chart resolution figures are becoming more and more meaningless, even in the digital era.

I think my eyes are bleeding......

Mission accompilshed

I believe his numbers are relevant

They aren't.

My layman's understanding (I am not a physicist, and my termonology may be incorrect) is that you can get incredibly high resolutions off a lens if you ignore the fact that eventually it hits something (either film or sensor), and that real world resolution numbers are always a fraction of theoretical maximized resolution numbers.

Why should this be? There are plenty of lenses (say kit lenses) which deliver EXACTLY their theoretical maximized resolution number into the final photo (which is why they can look awful). Why should this change with high end glass?

Resolution charts used to be published with both 1000:1 and 1.6:1 figures, and the lower contrast numbers were always significantly lower (generally ~50%) of the high contrast numbers.
One question that I have never recieved a satisfactory answer to is: Are optical sensors (DSLR sensors) prone to the same resolution losses as film in lower contrast pictorial use?
My gut says no, but......

Resolution is measured by defining some output contrast (e.g., 50% or 5%). So obviously, it drops when the input contrast is reduced. It would actually drop to zero if input contrast would be 30% but you require a 50% output contrast... But as I said, output contrast can easily be digitally enhanced if noise is low enough. So, low iso digital sensors have an edge here, if digital post processing is used.

[Wheatfield]

Thanks for taking the time to explain this. I didn't understand more than about 10% of it, but I appreciate your attempt.
I think what I got stuck on was Kodak's resolution numbers for T-Max 100.