MegaPixel frenzy vs. Lens capability

[KimH]

I read through the last couple of days of back and fro on 5D"?" and the lust for MP.

Someone mentioned that there are "less than a handfull" of current Canon lenses that can handle 64MP (someones wild shot), which makes me think and ask:

What is the optimal (or even a sensible) MP resolution on a FF to fit/match the good 70-200-II or the 4LIS, the 100 2.8L Macro.

To me it would appear that a 5DII has a sensor resolution max of 3800 LW/PH (Source PZ Klaus Schroiff) and these 2-3 great pieces of optics does not use the "space" allowed. Is it because they can't

The only one I have seen that might, is the 24mm 1.4-L-II (Centre only), I assume the 300 2.8LII and it's bigger sister would maybe...

Do i read this LW/PH all wrong, are there other limitation (MP or lens)?

Is there a sensible stream on this topic already or does anyone have a technical answer - just to give me some solace, while i wait for Friday
 

[Rav]

Going by the photozone tests, the 5DII is a good match for the 70-200 f4 IS, while the f2.8 IS II could use slightly more resolution at it's best setting.

As for other limiting factors, diffraction starts to be an issue the higher your sensor density is.

On a sidenote, when comparing resolution charts, be careful not to mix LW/PH and LP/PH. The 5DII with 3744 pixels height will never ever resolve more than 3744 LW/PH and consequently not more than 1872 LP/PH.

[Radiating]

I've actually done extensive testing to determine the resolution capabilities of Canon's current lens line up.

Most of their lenses are only capable of about 14 megapixels of resolution. Their L lenses tend to be closer to 17 and handful of lenses such as the 70-200mm F4 IS can go all the way up to 36 megapixels based on extrapolating results from crop cameras.


Here's the weird thing about resolution though, just because a lens can only render 14 megapixels, doesn't mean it needs a 14 mp sensor.

The best way to illustrate this is to go into photoshop create an alternating black and white set of lines. Then resize it using nearest neighbor to 10x it's size. Offset it by just 2 or 3 pixels and then resize it using bicubic to the original size.

What you did there is took an image of a certain resolution and did the equivalent of trying to render it at it's own resolution while inserting random variation. In the real world a len's image does not line up perfectly with a sensor's pixels. Lens tests compensate for this by measuring at hundreds of points. The reality is though that if you had 1 pixels for every point a lens could render you'd only capture 50% of the len's resolution.

Thankfully we have sharpening algorithms which help to even out the issue, which theoretically can halve the issue, unfortunately cameras use bayer sensors which make the problem slightly worse. In the end if you had 1 pixel for every point a lens could render on our cameras you'd only capture around 70% of it's resolution after processing.

The theoretical point of deminishing returns when trying to render a lens follows this graph, as mentioned earlier sharpening algorithms make the problem about half as bad, so this assumes there is zero sharpening:



Theoretically if you used a 14 megapixel lens on a 14 mp sensor and compared it to the same lens on a 28 mp sensor (double the resolution of the lens) you'd see a 21% increase in resolution, a 56 mp sensor (quardruple) would result in a 32% increase in resolution over 14 mp, and a 112 mp sensor would get you a 37% increase. 

So there is some benefit to having way more resolution than your lens can render simply to catch all the contrasting edges which do not perfectly align with the pixels on your camera.

[Ricku]

Is there any official info on how much MP each lens can handle?

I'd like to know how much MP the 135L, 35L and 70-200L II can take, without the IQ going downhill.

[te4o]

@Radiating: Thank you for the detailed and logical explanation (you must be a working physicist or similar to have a scientific approach like this - Max-Planck?) I have been asking the forum on this topic but just now got the explanation ! Thank you.
This info confirms that Canon and Co will continue gradually increasing the MP count of their sensors and, in parallel, gradually increasing the LP/PH of their lenses and we will be gradually paying more for both products if we want to use them together. 

[hendrik-sg]

@Radiating

Thanks for the analyses. What do you think of this thought?

- your analyses is roughly spoken a multiplication of two limitations (sensor and lens) which are applied to a picture with infinite contents (if i understand you correctly)

- Each picture has its own limitation due to discrete caracter of light with minimunm light quantity of one photon. The randomized distribution of the photons gives another limitation of resolution, given the amount of available light and observation time. That means a "perfect" lens/sensor combination can deliver the reals distribution of all photons.

- This limitation of the real would ad another multiplication of a similar curve like the ones you already multiplikated to the picture. In reality the physical limitation of the available information would flatten your curve even more and reducing the benefit of additionsal pixels even faster.