"Wasted" resolution of an "L" lens on an APC body??

I first want to thank everybody for replying with such an outstanding and informative set of replies. I have learned a lot here (and realize how poorly I understand all of this!).

My practical reason for posing this question originally was the basic confusion I had as to whether buying the 24-105L would gain me anything at all in resolution on my T1i when compared to the option of buying the new 15-85.

I totally understand that these are 2 entirely different lenses, and there are good reasons other than resolution to prefer one to the other. Among these differences are range of zoom, build quality, dust sealing, moisture sealing, vignetting, speed, weight, etc., so I did not have any confusion on these points. It merely seemed like a "waste" of the L lens to use it on an APC sensor body like my T1i, since the "system" resolution of both the lens and sensor as compared by photozone shows the 15-85 to be a bit better at the "best" choice of focal length and aperture from each of the 2 lenses. This was a surprsing comparison to me, since I assumed the L was far better in resolution.

I am still puzzled by the comment that these resolutions are not the key to selecting a sharp lens since the printed result will not reveal this resolution difference. I understand that even a big 11 by 14 print would not allow anywhere near the number of line pairs (2500 or so) to be visible, but certainly this resolution must direcly measure how well or poorly the detail in the original scene is captured. Cropping and pixel peeping certainly would reveal the difference, as would a print of an enlarged crop. Isn't this true? And thus the MTF/resoltution must be the key to knowing how well the lens resolves detail, even if it is not visible under some printing or other circumstances.

Thanks again for guidng me here.

Larry
 
...but certainly this resolution must direcly measure how well or poorly the detail in the original scene is captured.
You would think that, and if these were lens tests it would be true. But these tests are not lens tests, they're system test. System resolution is calculated:

1/Rs^2=1/Rc^2+1/Rl^2 where s is system, c is camera and l is lens.

The number you get from the PZ and DPReview tests is Rs. The problem is, Rl is usually much, much greater than Rc which means only a tiny fraction of Rl contributes to Rs while most of Rs is composed of Rc. Also, don't forget that there is measurement error, and that error can easily be on the same order as the contribution of Rl to Rs! What that means is that measurement error can easily be as important as differences in lens resolution when testing sharp lenses the way DPReview and Photozone test them, i.e. by looking through a blur filter (the AA filter) during the tests.

--
Lee Jay
(see profile for equipment)
 
The APC sensor uses the sweet spot of the lens so the lens will perform better than on a full frame camera...
This is just not true at all. I own both, do you?

Here's a sample of just how wrong you are:



--
Lee Jay
(see profile for equipment)
 
LSHorwitz1 wrote:

This was a surprsing comparison to me, since I assumed the L was far better in resolution.

An L lens designation does not guarantee higher resolution. It's usually the case, but some non L lenses have very high resolution (100mm macro), while some others don't (28-300 L). Canon also uses other criteria to determine L lenses. And of course there are no EF-S L lenses, and some of them are very sharp.
I am still puzzled by the comment that these resolutions are not the key to selecting a sharp lens since the printed result will not reveal this resolution difference. I understand that even a big 11 by 14 print would not allow anywhere near the number of line pairs (2500 or so) to be visible, but certainly this resolution must direcly measure how well or poorly the detail in the original scene is captured. Cropping and pixel peeping certainly would reveal the difference, as would a print of an enlarged crop. Isn't this true? And thus the MTF/resoltution must be the key to knowing how well the lens resolves detail, even if it is not visible under some printing or other circumstances.
The test results do measure how much detail is being captured by that particular lens/camera combination (subject to test conditions/errors). And the differences can be seen in larger prints.
http://www.ronhartman.net
 
In fact the 17-55 is sharper and has less distortions than the 24-70 but the 24-70 has much better color and bokeh...
Can you post some direct comparison tests-- same scene shot with both lenses?

I don't have the 24-70 to make a direct comparison, but looking at the photographs in the Landscape thread taken with the 24-70mm, I don't see anything different from the colors/quality I get with the 17-55mm.

regards,

-rich

--
Careful photographers run their own tests.
 
I fully appreciate that the comparison of lenses such as the 2 specific choices I selected (15-85 versus 24-105L) has many different dimensions / discriminants, and color, bokeh, weight, size, contrast, cost, and many other factors are all involved. My original question in this post and in the follow-up comments is, however, entirely related to resolution, and in this regard, resolution specifically, the "L" lens does seem to suffer a bit when used on an APC body. Hence my concern about "wasted" performance of the L lens, which does appear to work much better on a full frame body.

Larry
 
Consider what the test results say - lines/picture height rather than lines/mm which is the more scientific measure.

The full frame image is larger so, assuming image resolution is constant across the frame, its resolution should be a precise multiple of the measured resolution for the APS-C camera.

2500 lines for APS is the equivalent of 4000 lines for full frame (x1.6) so if anything, the FF camera is giving the anomolously low result. This is probably due to variation in pixel density or AA filter strength.
 
A simple way to think about this is as follows:

A lens can resolve X lp/mm at the surface of any sensor. However different size sensors can contain different number of line pairs, with the larger sensor holding more of them.

An important value in your consideration of two formats is line pairs per picture width rather than purely lp/mm.

Dan

--
---
G Dan Mitchell - SF Bay Area, California, USA
Blog & Gallery: http://www.gdanmitchell.com/
Facebook: http://www.facebook.com/gdanmitchell
Flickr: http://www.flickr.com/photos/gdanmitchell/
Twitter: http://twitter.com/gdanmitchell
IM: gdanmitchell

Gear List: Cup, spoon, chewing gum, old shoe laces, spare change, eyeballs, bag of nuts.
 
You used lenses with big enough sweet spots to essentially cover the complete area of a full frame sensor. The results would be very different if you used the 17-40L lens which suffers from soft corners even at f8 and has no corner resolution at all at f4.
The APC sensor uses the sweet spot of the lens so the lens will perform better than on a full frame camera...
This is just not true at all. I own both, do you?

Here's a sample of just how wrong you are:



--
Lee Jay
(see profile for equipment)
 
The picture height is 1.6 bigger for full frame cameras. If you are interested in the performance in crops you need to divide the full frame resolution by 1.6 to get comparable results. The results is that the lens has noticeably higher resolution on the denser APS-C sensor.

If you are interested in complete images instead of small areas the full frame sensor will work better with a low resolution lens as long as that lens does not suffer from corner softness.
I am trying to understand how a specific Canon lens behaves on an APC versus a full frame body, and have been looking at the photozone.de comparisons of the Canon 24-105 L lens specifically.

The specific performance of the lens I am trying to compare is resolution. The photozone.de analysis of this lens for the APC sensor shows that the lens peaks at around 2500 line pairs per page height at its best performance. See the tables at:

http://www.photozone.de/canon-eos/423-canon_24105_4_50d?start=1

The identical lens on the full frame body shows around 3400 line pairs per page height when evaluated on a full frame body by photozone. See:

http://www.photozone.de/canon_eos_ff/420-canon_24105_4_5d?start=1

I understand that the smaller APC sensor receives a "cropped" view of the lens output, when compared to the full frame sensor. I would therefore have assumed that the smaller APC sensor should resolve, at best, 66% of the line pairs seen by the full frame. Instead it is apparently able to achieve about 75% of the resolution.

More important is the issue of whether a body such as Canon's T1i, which supposedly allows 3200+ lines of resolution, will "waste" the real benefit of an "L" lens since it apparently cannot see/use the remaining performance of the lens.

Is there an 'optimal' resolution for best achieving sharpness with an APC sensor given the above, or does the "L" lens really enjoy some type of advantage over a non-L lens even if used on an APC camera.

I specifically will note that the new Canon 15-85 non-L zoom lens, for example, appears to offer HIGHER resolution peaking at 2548 line pairs when compared to the 24-105mm L lens, and thus could imply that it is possibly a BETTER match for the APC sensor.

Am I understanding all of this correctly, or do I have some basic misunderstandings here?

Many thanks for comments.

Larry
 
You used lenses with big enough sweet spots to essentially cover the complete area of a full frame sensor. The results would be very different if you used the 17-40L lens which suffers from soft corners even at f8 and has no corner resolution at all at f4.
Sorry, no, I've done that test too.

17-40L f4 20D 17mm versus 17-40L f6.3 5D 27mm, same FOV, same total light, same DOF, 5D wins.

--
Lee Jay
(see profile for equipment)
 
Sorry, no, I've done that test too.

17-40L f4 20D 17mm versus 17-40L f6.3 5D 27mm, same FOV, same total light, same DOF, 5D wins.
The key here is:

same_FOV

This is the central point of contention between differing sensor sizes. When zooming is possible (either via the lens or via feet), the larger sensor will always come out on top, eveything else equal. However, for fixed positions and fixed focal lengths of the lens, the higher pixel density (usually smaller) sensors will always come out on top, assuming the subject fits within the FOV of the smaller sensor.

A good example are the longer telephoto primes. If one is "reach"-limited (i.e. - you have 'only' a 400mm lens, and not a 600m or 800mm lens in your bag), then the cropped sensor with its higher pixel density will produce a sharper image than the larger sensor with a lower pixel density, simply because you will have to crop the FF picture. This is why many/most birders and 'stadium' sports photogs prefer APS-H or APS-C over FF and why - to the chagrin of many - Canon released the 1D MkIV in APS-H format rather than FF format.
 
However, for fixed positions and fixed focal lengths of the lens, the higher pixel density (usually smaller) sensors will always come out on top, assuming the subject fits within the FOV of the smaller sensor.
That's not what we were talking about, but it's also true - higher pixel density wins when you are focal length or magnification limited. Here's an example, 20D on top, 5D on the bottom, both at the same focal length, both 100% crops:



--
Lee Jay
(see profile for equipment)
 
Good morning Lee Jay,

Are you sure that the resolutions of lens and sensor combine with squared terms in the denominators? You're the only one I've ever seen put it that way. What I have always thought is that they combined like resistors in parallel with first order terms. ie., 1/Rs = 1/Rc + 1/Rl.

That's what this guy thinks as well.
http://www.dphoto.us/forum/showthread.php?t=5612

Can you tell me how you arrived at the squared terms?
...but certainly this resolution must direcly measure how well or poorly the detail in the original scene is captured.
You would think that, and if these were lens tests it would be true. But these tests are not lens tests, they're system test. System resolution is calculated:

1/Rs^2=1/Rc^2+1/Rl^2 where s is system, c is camera and l is lens.

The number you get from the PZ and DPReview tests is Rs. The problem is, Rl is usually much, much greater than Rc which means only a tiny fraction of Rl contributes to Rs while most of Rs is composed of Rc. Also, don't forget that there is measurement error, and that error can easily be on the same order as the contribution of Rl to Rs! What that means is that measurement error can easily be as important as differences in lens resolution when testing sharp lenses the way DPReview and Photozone test them, i.e. by looking through a blur filter (the AA filter) during the tests.

--
Lee Jay
(see profile for equipment)
--
kind regards
Dale
 
why 40D/50D has a higher diffraction limits being a body with higher resolution ?

thanks
Further, diffraction limits set in on the 7D at f/11. The 5D2 diffraction limits at f/16. The 40D and 50D limit somewhere in between.

--
Christopher Perez
  • PhotoLife Magazine November 2009 - Steampunk
  • LensWork Extended #78 Sept-Oct 2008 - In the Railyard
  • CameraWork Gallery 2007 - Palladium prints
  • The sharpest lens in my kit is a tripod
  • Cameras have no talent. Its what's behind the eyepiece that counts.
--
http://www.holeepassion.com .. photo gallery coming soon
 
Very interesting reading Lee jay. I noticed that these links you included are the posts where you had a change of heart, and changed the formula yourself from fist order to second order terms. I also noticed that one of your favorite quotes prior to that was from Chuck Westfall, who was using the first order equation.

I'm not totally converted myself yet, but I am leaning that way based upon the arguments presented in your links and some of my own calculations.

For example: Assume the both the sensor and the lens have equal resolutions of say 1000 lpm. Using 1/Rf=1/Rs + 1/Rl - then 1/Rf - 2/1000 or 500 lpm. On the other hand using the squared term equation that Kodak likes, and the same sensor and lens resolutions, the Rfinal equals 707 lpm, which frankly seems more reasonable to me. So that's where I'm at... probably where you were a while ago. :)

Both equations give results that many users here don't anticipate however, and that is that when one says "the sensor out-resolves the lens". They think that's a bad thing. It's not. That would be a very good thing. It allows you to get closer to what the lens' true potential is.

It's the same way with very fine-grained high resolution film. It doesn't make the lens give you worse resolution, it always gives better results than film with larger grain. The same with digital sensors. They're like fine grained film.

Thanks for the heads up.
Good morning Lee Jay,

Are you sure that the resolutions of lens and sensor combine with squared terms in the denominators? You're the only one I've ever seen put it that way.
http://forums.dpreview.com/forums/read.asp?forum=1032&message=21434573
http://forums.dpreview.com/forums/read.asp?forum=1029&message=24585118

--
Lee Jay
(see profile for equipment)
--
kind regards
Dale
 
Very interesting reading Lee jay. I noticed that these links you included are the posts where you had a change of heart, and changed the formula yourself from fist order to second order terms.
Correct. I never derived it so I never had any reason to defend it. Others claim to have derived it and gave a reasonable heuristic method of understanding. That said, I still haven't derived it so I still can't defend it with authority.

--
Lee Jay
(see profile for equipment)
 
Very interesting reading Lee jay. I noticed that these links you included are the posts where you had a change of heart, and changed the formula yourself from fist order to second order terms.
Correct. I never derived it so I never had any reason to defend it. Others claim to have derived it and gave a reasonable heuristic method of understanding. That said, I still haven't derived it so I still can't defend it with authority.
Sigh...

http://www.normankoren.com/Tutorials/MTF.html

"Resolution of an imaging system (old definition)— Using the assumption that resolution is a frequency where MTF is 10% or less, the resolution r of a system consisting of n components, each of which has an MTF curve similar to those shown below, can be approximated by the equation, 1/r = 1/r1 + 1/r2 + ... + 1/rn (equivalently, r = 1/(1/r1 + 1/r2 + ... + 1/rn )). This equation is adequate as a first order estimate, but not as accurate as multiplying MTF's. [I verified it with a bit of mathematics, assuming a second order MTF rolloff typical of the curves below. It's not sensitive to the MTF percentage that defines r. The approximation, 1/r2 = 1/r12 + 1/r22 + ..., is not accurate.]"

http://en.wikipedia.org/wiki/Optical_resolution#System_resolution

That has the correct equations which don't involve a resolution for a particular MTF. I guess either the 1/r or 1/r^2 approaches are just approximations that depend on a particular PSF (Gaussian) which isn't really true in real life. The 1/r^2 thing appears to be correct for that case, but not for real systems like AA filters and spherical aberration.

--
Lee Jay
(see profile for equipment)
 
Thanks for the update, Lee Jay. Very good stuff.

It seems there is a lot going on to get the "real" system resolution with lens, sensor, and AA filter including all optical effects. The math gets very nasty with too many variables.

But, I think the second order terms give what to me seems like more reasonable results than the first order approximation.

In any event, both equations show that more sensor resolution is a good thing for the overall system resolution, even if the sensor out-resolves the lens .. which many have been saying, based upon empirical results, for some time now. However, there are still naysayers.
Very interesting reading Lee jay. I noticed that these links you included are the posts where you had a change of heart, and changed the formula yourself from fist order to second order terms.
Correct. I never derived it so I never had any reason to defend it. Others claim to have derived it and gave a reasonable heuristic method of understanding. That said, I still haven't derived it so I still can't defend it with authority.
Sigh...

http://www.normankoren.com/Tutorials/MTF.html

"Resolution of an imaging system (old definition)— Using the assumption that resolution is a frequency where MTF is 10% or less, the resolution r of a system consisting of n components, each of which has an MTF curve similar to those shown below, can be approximated by the equation, 1/r = 1/r1 + 1/r2 + ... + 1/rn (equivalently, r = 1/(1/r1 + 1/r2 + ... + 1/rn )). This equation is adequate as a first order estimate, but not as accurate as multiplying MTF's. [I verified it with a bit of mathematics, assuming a second order MTF rolloff typical of the curves below. It's not sensitive to the MTF percentage that defines r. The approximation, 1/r2 = 1/r12 + 1/r22 + ..., is not accurate.]"

http://en.wikipedia.org/wiki/Optical_resolution#System_resolution

That has the correct equations which don't involve a resolution for a particular MTF. I guess either the 1/r or 1/r^2 approaches are just approximations that depend on a particular PSF (Gaussian) which isn't really true in real life. The 1/r^2 thing appears to be correct for that case, but not for real systems like AA filters and spherical aberration.

--
Lee Jay
(see profile for equipment)
--
kind regards
Dale
 

Keyboard shortcuts

Back
Top