Cureent Canon lenses vs resolving power of sensor

monky said:

i guess then my issue is that i find it odd to use a figure of
100lp/mm when the best lenses are already down to 50% at 45lp/mm.
by 100lp/mm those lenses will be around 10%... to me, that doesn't
seem like a limit to work towards for the sensor when the lenses
are already severly limited in microcontrast in that range.

Click to expand...

According to Castleman, the 70-200 2.8L IS has a 50% MTF of about 27lp/mm at f5.6:



The 20D has 6.4 micron pixels which, if you include the Bayer mask, Kell factor and processing, gives you a resolving power (at 3 pixels per lp) of 52lp/mm. So the camera already out-resolves the lens by quite a bit, right?

If that's true, adding teleconverters to that lens will not provide any additional image information since the sensor is already extracting it all. Therefore up-resing will be as good as, or better than adding the TCs. You be the judge:



There's a lot of very usable available resolution past 50% MTF.

--
Lee Jay
(see profile for equipment)

I know that adding another 1.4 to my 70-200 2.8L IS provides very little useful additional information over the 2x. So that indicates that that lens provides 104lp/mm (or a little more) of usable resolution.

That's from an L-zoom! The L-primes are better. Several people have shown conclusively that the 300/2.8L IS is able to provide more information to the 20D when used with stacked 1.4x and 2xTC than with a 2x alone. That fact indicates that that lens can provide at least 146lp/mm of useful resolution (2*1.4*52). That corresponds to a 1.6 crop sensor of about 65MP.

--
Lee Jay
(see profile for equipment)

monky said:

this is the MTF for the 24L. the 50% MTF should be almost exactly
30lp/mm at f/1.4. this is pretty much in line with castleman's
findings... around 25lp/mm.

Click to expand...

I'm sorry, I don't follow...

monky said:

i guess then my issue is that i find it odd to use a figure of
100lp/mm when the best lenses are already down to 50% at 45lp/mm.
by 100lp/mm those lenses will be around 10%... to me, that doesn't
seem like a limit to work towards for the sensor when the lenses
are already severly limited in microcontrast in that range.

Click to expand...

Essentially, it doesn't really matter whether a lens can reproduce a 100lp/mm pattern at 50% MTF or 2% MTF - the fact that it can reproduce it at all is enough Here is a visual representation of MTF at different spatial frequencies:



As you can see, even at 2% MTF there is a discernible pattern (IOW the lens was capable of resolving the line pairs) and so if this detail, however poor, is not to be lost completely then the sensor also has to be able to resolve it.

Steve H
--

but what is a resonable limit on the MTF to use as a constraint to ask the question "where does the lens limit the process"?

is there a way to decide exactly where the MTF is at 104 lp/mm (using your example)?

and thanks for showing how a lens can exceet 100lp/mm, but this would also imply that tech pan is what was limiting castleman's tests, correct?

--



ed murphy ----------- AIM: monky9000

monky said:

but what is a resonable limit on the MTF to use as a constraint to
ask the question "where does the lens limit the process"?

Click to expand...

The lens always limits the process. So does the sensor. The MTF of the final image is a combination of the MTFs of both parts. That's why the 150lp/mm 300/2.8 looks sharper than the 104lp/mm 70-200 2.8L IS even without TCs.

monky said:

is there a way to decide exactly where the MTF is at 104 lp/mm
(using your example)?

Click to expand...

I don't think so. I'd guess in the 20%-30% range though this is just a guess.

monky said:

and thanks for showing how a lens can exceet 100lp/mm, but this
would also imply that tech pan is what was limiting castleman's
tests, correct?

Click to expand...

I think the combination of the techpan (which is great stuff), the 50% limit and the scanner made up that result.

--
Lee Jay
(see profile for equipment)

Hey Lee and anyone else who might be as knowledgeable as Lee,

What do you think is the reason for the IS to win in MTF-50 but the non-IS to clearly win in max-MTF?





Don't MTF-50 and max-MTF usually have a correlation, or are there really lenses showing them to be completely independent? I suppose the coatings and microstructure of the glass could show a preference for passing certain frequencies, but it would just seem that "good glass" would normally have a good all-around frequency response. Still, the book EF Lens Works III (I think that's the one) comes to mind, where it shows examples of high contrast with low resolution vs. high res w/low contrast and then hi/hi. But are there really lenses that yield such a disparity?

Also, why do you think the PhotoZone tests found the non-IS to clearly beat the IS in MTF-50? (See http://www.pbase.com/lwestfall/image/51256678 ) Just sample variation?

ljfinger said:

According to Castleman, the 70-200 2.8L IS has a 50% MTF of about
27lp/mm at f5.6

Click to expand...

But isn't that the MTF-50 of the SYSTEM, not just the LENS? ( Along these lines: http://www.pbase.com/lwestfall/image/52085939 )

BTW, in post processing, isn't it easier to "recover" (i.e. mask or make up for) low contrast than low resolution? Or is the RAW dynamic range given by current Canon DSLRs not good enough to make this statement?

Sorry to be asking so many questions all at once that may not have clear answers.

Thanks,
Lincoln

--
ALL PhotoZone.de lens test data compared graphically:
http://www.pbase.com/lwestfall/lens_tests

gml said:

I don't have any scientific data to prove it but my observations
are that the 20D already outresolves most lenses. So, I’d say that
the maximum meaningful resolution is 8-10MP for 1.6x sensors and
20-24MP for FF sensors. Increasing sensor resolution beyond that
will not result in improved image quality.

Click to expand...

I don't know whether we can say "outresolve" there, but I agree that the 20D, with its high pixel density, seems to make sharp lenses look soft at the pixel level. However, my understanding is that your SYSTEM's "meaningful resolution" does indeed continue to increase whether you increase LENS resolution or SENSOR resolution. See http://www.pbase.com/lwestfall/image/52085939

Lincoln

--
ALL PhotoZone.de lens test data compared graphically:
http://www.pbase.com/lwestfall/lens_tests

In the old days, when MODERN PHOTOGRAPHY published lens tests, there were a few lenses (I remember Leica, Alpa, Nikon, Vivitar Series 1 Macro) that surpassed 100 Lp/mm. Generally, thei "excellent" ratings applied to lenses above 80 Lp/mm.

But that was in the sixties and seventies. Today's L lenses may well be better.
--
Fritz

http://www.pbase.com/fwscharpf/galleries

for a full-frame sensor....... these are not my calculations .... but those of others that I have trawled from this site and the Web. The problem is that while image size/quality is increasing, the technology to store these images (e.g. CF cards) is not and is (artificially?) expensive. It's time for someone to market a camera that transmits it images to a separate large capacity hard drive that you can keep in your pocket ...... or why not incorporate a hard drive into an accessory grip?

insectman said:

The problem is that while image size/quality is increasing, the technology to store these images (e.g. CF cards) is not ..

Click to expand...

Not sure what you mean here.... do you mean the size of the cards?

A couple of years ago a 512MB CF card was absolutely massive. Now you can buy 4GB CF cards quite easily.

So the size of the cards is surely increasing quite rapidly. Or did you mean something else?

lwestfall said:

Hey Lee and anyone else who might be as knowledgeable as Lee,

What do you think is the reason for the IS to win in MTF-50 but the
non-IS to clearly win in max-MTF?

Click to expand...

I don't know.

lwestfall said:

Still, the book EF Lens Works III (I think that's the
one) comes to mind, where it shows examples of high contrast with
low resolution vs. high res w/low contrast and then hi/hi. But are
there really lenses that yield such a disparity?

Click to expand...

Sure. Contrast can be reduced by internal lens reflections but resolution is reduced (largely) by spherical aberration. Those are pretty independent.

lwestfall said:

Also, why do you think the PhotoZone tests found the non-IS to
clearly beat the IS in MTF-50? (See
http://www.pbase.com/lwestfall/image/51256678 ) Just sample
variation?

Click to expand...

Probably.

lwestfall said:

ljfinger said:

According to Castleman, the 70-200 2.8L IS has a 50% MTF of about
27lp/mm at f5.6

Click to expand...

But isn't that the MTF-50 of the SYSTEM, not just the LENS? (
Along these lines: http://www.pbase.com/lwestfall/image/52085939 )

Click to expand...

Yes. The idea of using TechPan in Technidol is to make the film much, much better than the lens. In reality, the combination is still the right answer.

lwestfall said:

BTW, in post processing, isn't it easier to "recover" (i.e. mask or
make up for) low contrast than low resolution?

Click to expand...

Absolutely it is.

--
Lee Jay
(see profile for equipment)

The effective resolution, Re, which is the point at which the MTF falls to about 5% or so, reflects the resolution of both lens, Rl, and sensor. Rs.

Re is roughly 1/(1/Rs + 1/Rl). The result is that you always benefit a lot by improving either unless one totally dominates.
eg.
Rs=50lp/mm; Rl=50lp/mm => Re=25lp/mm
Rs=100lp/mm; Rl=50lp/mm => Re=33lp/mm
Rs=200lp/mm; Rl=50lp/mm => Re=40lp/mm

but it is clearly diminishing returns.

I got the above relationship from:
http://www.normankoren.com/Tutorials/MTF.html#Human_visual_acuity

--
Charlie

Better lucky than smart. The more shots I take, the luckier I get.

monky said:

how did you arrive at 24MP?

Click to expand...

The follwing page shows a 20D resolving at the 62 lp/mm to 71 lp/mm level (depending upon how resolution is measured). Say the 8.2 MP 20D resolves at 62 lp/mm, then to obtain 100 lp/mm one would need 1.62*1.62 times as many pixels or 21.3 MP so I rounded up to 24MP to be generous.

http://www.wlcastleman.com/equip/reviews/film_ccd/index.htm

monky said:

i am getting the same number steve arrived at (29.6 actually),
assuming i use the '3' for an interpolated bayer sensor insead of
the customary '2' for nyquist (with which i arriave at just 13.1MP).

Click to expand...

The green channel (running diagonally) has a resolving power of 1/SQRT(2) = 1/1.414 of the pixel spacng.

monky said:

and this is all assuming i use 100lp/mm of course, and i am still
wondering exactly where that number came from.

Click to expand...

Only the very best camera lense resolve on the very best of film at a level higher than this, and only in a very narrow range of relative apertures. Most lenses do not resolve at this level. On camera lense I hae not seen any resolution nubers above 120 lp/mm and very few in the 110 lp/mm level.
--
Mitch

lwestfall said:

gml said:

I don't have any scientific data to prove it but my observations
are that the 20D already outresolves most lenses. So, I’d say that
the maximum meaningful resolution is 8-10MP for 1.6x sensors and
20-24MP for FF sensors. Increasing sensor resolution beyond that
will not result in improved image quality.

Click to expand...

I don't know whether we can say "outresolve" there, but I agree
that the 20D, with its high pixel density, seems to make sharp
lenses look soft at the pixel level. However, my understanding is
that your SYSTEM's "meaningful resolution" does indeed continue to
increase whether you increase LENS resolution or SENSOR resolution.
See http://www.pbase.com/lwestfall/image/52085939

Click to expand...

I've seen the graphs before. System resolution is indeed a factor of sensor and lens resolution. However, system resolution is actually limited by the max resolution of either the lens or the sensor. For example, let’s say you have a 2mp sensor. According to your formula, system resolution will keep increasing as you increase lens resolution. In practice, however, you won’t be able to resolve more detail in a scene than a 2mp sensor can resolve, regardless of how much you increase lens resolution. Or, let's say you have 48gp (giga-pixel) sensor - it will resolve only as much detail as the lens resolves. You can argue that with a 48gp sensor your SYSTEM has a huge resolution but that’s actually misleading, since it obviously doesn't apply to system's ability to resolve detail in a scene.

George

I'm glad there are people like you to debate this stuff. It makes no sense to me. Seeing my images in magazines lets me see how sharp they are...not graphs

Paul

This has been an extremely informative thread. Thanks everyone !!
--
I am always learning something here on the forums.

Super answers.

Just a comment about sensor resolution.

The pixel resolution is based on linear pixels per millimeter of the sensor, but isn't this an RGB stripe (or whatever) pattern?

How does this affect the true resolution?

My theory on how the pixels are resolved as full colour 24 bit per pixel when each is actually one of R, G or B is a bit sketchy.

Perhaps for black and white MTF tests each colour filtered pixel helps to resolve the edges of the black/white divide, and so the resolution for these tests really is the MP resolution?