Jump to content

MTF Curves ~ Sensor Resolution


k-hawinkler

Recommended Posts

Advertisement (gone after registration)

It does? So does mine :rolleyes:

 

 

 

Oh dear. Where to start? Both you and 'k-hawinkler' are badly confusing things like lines, line pairs, Bayer patterns, Nyquist-Shannon, and a few more and mixing them up into a melange of myths, factoids, and misconceptions which is just short of impossible to sort out.

 

Among other things, line pairs are not lines. The Nyquist-Shannon theorem does not say 'at least two samples per cycle;' actually it says, 'more than two samples per cycle.' And 1.5 is not shorthand for 'somewhere between 1 and 2.' The number quoted was '1.4 or 1.5,' and that's what it meant.

 

 

 

Divide these results by two, then you're closer to what we actually get—in theory, in the general case (special cases not withstanding). After all, it's line pairs, not lines. This will be still not necessarily the same what we get in real life, because what we really get depends on many factors, including (but not limited to) the minutiae of the in-camera image processing as well as contrast, orientation, spatial phase, and colour of the structures to reproduce.

 

And then, that's only what the sensor can do. This still is different from what you'll see in an image cast by a lens on the sensor. For the image resolution to approach the sensor's resolution, the lens must approach infinite resolution ... which is impossible obviously. So image resolution will be always less than the sensor's theoretical limit, and there is no such thing as a lens that 'outresolves' a sensor.

 

 

Hi 01af,

 

Thanks for your interest in this thread and your feedback. I appreciate it.

 

Sometimes, though, you seem to be barking up the wrong tree.

You might want to consider honing your communication skills.

 

For example, I don't think I characterized or offered an opinion on the concepts you mentioned in the first two paragraphs.

Just because I list a reference doesn't mean I agree with or endorse it. However, I am looking for feedback from forum members.

I am not asking rhetorical questions that I already know the answer to, but questions about subjects that interest me.

 

I also would like to suggest you might consider following up your sweeping characterizations and condemnations of Do Sensors “Outresolve” Lenses? with some specific examples of errors. Otherwise I have no clue what you are talking about.

 

Thanks for the last two paragraphs. I think I got it.

Link to post
Share on other sites

01AF can indeed be a bit brusque, but his problem is that there is vast oversimplification both in the questions asked and in the LL article. To counter that in this very convoluted subject as forum post is far to small. It needs a book, or at least a long in-depth article...

Link to post
Share on other sites

01AF can indeed be a bit brusque, but his problem is that there is vast oversimplification both in the questions asked and in the LL article. To counter that in this very convoluted subject as forum post is far to small. It needs a book, or at least a long in-depth article...

 

 

Well, my expectation of forum members always is that they express themselves in a precise and respectful manner. Of course, all of us are humans ...

Link to post
Share on other sites

K-H,

 

There is a decent summary of this entire topic at the following link

Norman Koren photography: images and tutorials

Look under his tutorials on MTF, sharpness, etc. He does decent explanation and has lots of plots to illustrate the discussion. His site has not been updated in many years, so the discussion of digital sensors is a little outdated but that doesn't matter for this discussion. That reading should keep you busy for the rest of the day....

 

Regards,

 

RM

  • Like 3
Link to post
Share on other sites

It does? So does mine :rolleyes: ...... The Nyquist-Shannon theorem does not say 'at least two samples per cycle;' actually it says, 'more than two samples per cycle.'

 

Sorry to make your head ache again ;) - but can you show a significant semantic distinction between "at least two samples per cycle" and "more than two cycles per sample"?

 

2 = "at least two..."

2.00000000000.....001 = "more than two....."

 

Anyway, I agree with both your and jaap's constructions, even if you didn't like mine. Peace!

Link to post
Share on other sites

... can you show a significant semantic distinction between "at least two samples per cycle" and "more than two cycles per sample"?

The numeric difference may be infinitesimally small but the semantic difference is glaringly obvious.

 

 

2 = "at least two ..."

2.00000000000.....001 = "more than two ..."

Doh ... so you can see the difference. Then why are you asking about semantics in the first place? Two is not enough—as simple as that.

 

Mathematically, any amount beyond two will do. In real life, however, resources like time, space, and energy are not infinite so we have to deal with limited computing power, limited accuracy, and noise. So the number of samples per cycle better be somewhere between say, 2.5 and 3, or even higher still. In a picture, detail that is represented with, say, ten or twenty pixels will always look better, clearer, richer, sharper than detail at the very limit of what the sensor can just barely resolve.

 

 

There is a decent summary of this entire topic at the following link:

 

Norman Koren photography: images and tutorials

 

Look under his tutorials on MTF, sharpness, etc. He does decent explanation and has lots of plots to illustrate the discussion.

Wow! That's a great resource! After skimming quickly over this text (and following pages), I don't see any obvious mistakes. That's new. Several years ago, Norman Koren wrote the same nonsense about sensors, pixels, Airy disks, and resolution as anyone else. This is a typical topic where on the Internet one fool is copying from the next, and actual and accurate information from real experts is hard to find. I've been writing about it over and over but to no avail so far; the fan-out of the self-appointed half-educated 'experts' (like, for example, Nathan Myhrvold or Rubén Osuna etc.) simply blows away mine. And I am far from an expert either; I have just a outlined but at least correct understanding of the basics.

 

 

His site has not been updated in many years ...

Oh, it definitely has. At the bottom of the page it says, 'Text and photographs copyright © 1999-2012 by Norman Koren,' and the contents of the page about MTF and sharpness definitely has been updated in the last one or two years (even though parts of it are older).

Edited by 01af
Link to post
Share on other sites

Advertisement (gone after registration)

Sorry to make your head ache again ;) - but can you show a significant semantic distinction between "at least two samples per cycle" and "more than two cycles per sample"?

Suppose you have an image of black and white lines, each the width of a single sensor pixel. When this image is perfectly aligned with the pixel grid then each sensor pixel will see either a black line (and nothing else) or a white line (and nothing else). The result is a digital image with maximum contrast where the lines are perfectly resolved. But now assume that the alignment is such that each sensor pixel will see exactly one half of a black line and one half of a white line – in that case, each pixel sees a middle gray and the whole image will be gray, i.e. contrast equals zero and nothing is resolved. So if you have two pixels to resolve a line pair, contrast can be anything between maximum and zero, depending on the circumstances. But once the sensor pixels are only slightly smaller than the lines so that you have slightly more then 2 pixels per line pair, the contrast cannot be zero, even when it may still be small in the worst case.

 

Now in practice you are right in that there is nothing magical happening when you cross that ‘two samples per cycle’ line. Sampling with a slightly lower frequency will often yield acceptable images and if it doesn’t, sampling with a slightly higher frequency won’t result in dramatically improved results. But the sampling theorem is a mathematical theorem and in mathematics you don’t get away with ignoring the difference between > and ≥.

  • Like 8
Link to post
Share on other sites

 

Wow! That's a great resource! After skimming quickly over this text (and following pages), I don't see any obvious mistakes. That's new. Several years ago, Norman Koren wrote the same nonsense about sensors, pixels, Airy disks, and resolution as anyone else. This is a typical topic where on the Internet one fool is copying from the next, and actual and accurate information from real experts is hard to find. I've been writing about it over and over but to no avail so far; the fan-out of the self-appointed half-educated 'experts' (like, for example, Nathan Myhrvold or Rubén Osuna etc.) simply blows away mine. And I am far from an expert either; I have just a outlined but at least correct understanding of the basics.

 

[/color][/size][/font]

 

 

Yes, I can understand your frustration on reading unrefereed, internet articles.

If you want to read THE definitive text, go to a University Library and find Born & Wolf "Principles of Optics".

It is a testament to the quality of this work that it is still in print, has updated versions, and is still referred to as the definitive text(and reference). If I recall right, the first edition of this work predates Lasers. Yes, that is the same Max Born who helped invent quantum mechanics in the 1920's/30's and passed away over 40 years ago.

 

That text will keep you occupied for a semester, graduate level course in physics. So, if this discussion gives you a headache, be forewarned......

 

Regards,

 

RM

  • Like 1
Link to post
Share on other sites

I have been trying for some time to buy a digital copy of Born & Wolf "Principles of Optics".

None is for sale on amazon.com for example.

However, apparently there seem to be complete PDF files on the Internet for downloading.

I suspect those are in violation of copy right laws.

 

Question: Is there a digital copy available for sale or download that is in compliance with the laws?

I suppose not, but I am posing my question anyway.

 

Thanks.

Link to post
Share on other sites

K-H,

 

For a handy, pocket guide book size, you might also try the book:

"Field Guide to Geometrical Optics", J.E. Greivenkamp ISBN 0-8194-5294-7.

This is part of the SPIE Field Guide series. It is a handy, spiral bound book, soft cover, with useful summary information all in one place. You can probably find more info at spie.org or do a www search. Hope that helps.

 

RM

  • Like 1
Link to post
Share on other sites

  • 9 months later...
  • 2 weeks later...
Keep in mind that "line pairs per millimeter" is the same as "lines per millimeter," since it takes both black and white to make a single "line."

 

 

Hi Howard,

 

Thanks for the feedback.

I have not heard from you in awhile and hope you are doing fine.

It's always a pleasure to read your contributions.

 

Well, to me it doesn't matter how they count.

Of course, you need a black and a white line in a pair.

Some folks seem to count that as two lines.

 

Thanks again.

I will keep this in mind when researching the issue further. :)

Link to post
Share on other sites

A lens is an analog device and a digital camera is...well.....digital. You may have trouble getting a direct comparison.

 

 

Hi Ray,

 

I have learned a lot since I started this thread, but mostly independent of it.

The LensRentals reference above is very informative and sheds some light on how sensor and lens contribute to image quality, for example resolution. :)

Link to post
Share on other sites

Not to get into a tech discussion, all I can say is lenses that are already nearly perfect such as my 90 4.0 macro make a sharp image already at 4.0 and stopping down only improves depth of field, digital M9 or film.

 

A lens like the 200 4.0 is slightly soft on film or digital at 4.0. It sharpens to max at 5.6 on either medium. Same as other samples of same lens. Same with my 135 2.8 , but it is not the last version. Put it on a viso to eliminate possible focusing errors and there is no change. My Viso is calibrated to digital and film. The last version of the 135 2.8 is very slightly better at 2.8.

 

I have no way to calculate on the new M 240.

 

When I have used older viso lenses on a Nikon D3, 12 MP with AA filter, there is less difference at wide open and stopped down compared to film which suggests to me film out resolves digital set up this way. This also tells me the lenses out resolve a 12 MP sensor with AA filter, but not higher MP sensors without AA filter . The D3 focus is perfect as screen is matched to live view sensor and green focus dot. They do not come the factory calibrated this closely.

 

I am going to suggest Nikon D800 e will out resolve film and lenses with less than perfect resolution will show it wide open. Same with the M 240 and M9.

 

There is now a need for better lenses with the MM and and its high acuity sensor and perhaps the M 240. Hence we have the 50 2.0 APO. I bet there is a 35 MM APO is the works.

 

The whole imaging is a chain only strong as weakest link. Now sensors are better so new lenses will have to catch up.

  • Like 1
Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...