Sensor out resolving the lens or the opposite?

[Theodoros]

There seems that there are preferences on both "sensor out resolving the lens" and "lens out resolving the sensor" sides… which side are you on and (most importantly) why? It seems that MF makers tend to favour the first for their higher end models…, are they right in doing so?

[eronald]

There seems that there are preferences on both "sensor out resolving the lens" and "lens out resolving the sensor" sides… which side are you on and (most importantly) why? It seems that MF makers tend to favour the first for their higher end models…, are they right in doing so?

As soon as the lens out-resolves the sensor, theory predicts Moiré.
Which is exactly what happens on the older MF backs.

Edmund

[Theodoros]

As soon as the lens out-resolves the sensor, theory predicts Moiré.
Which is exactly what happens on the older MF backs.

Edmund

Moire seems to be a "thing of the past" as software develops to overcome it… It's a rare case too and can happen even with sensor out resolving the lens… So, despite moire, what should preferable?

[synn]

I prefer lenses that render a pleasing image and sensors that reproduce colors faithfully.
"Outresolving" is the last thing on my mind.

[Theodoros]

I prefer lenses that render a pleasing image and sensors that reproduce colors faithfully.
"Outresolving" is the last thing on my mind.

Certainly faithful colour reproduction is a sensor quality, but can you define "pleasing image rendering" more accurately?

[synn]

Certainly faithful colour reproduction is a sensor quality, but can you define "pleasing image rendering" more accurately?

It's a very qualitative thing. Also, somewhat subjective.
To give an example, I shoot Nikon for 35mm and while a lot of their lenses are clinically tack sharp and have through the roof ratings at DxO and what not, most of them render a very boring image. A few, like the 85 f/1.8G and the 200 f/2 G however, render images in a very pleasing way.

[ErikKaffehr]

Hi,

Just as it happens I have updated a small article I have written on the issue:

http://echophoto.dnsalias.net/ekr/index.php/photoarticles/78-aliasing-and-supersampling-why-small-pixels-are-good

I don't think raw processors can handle moiré, they can reduce colour moiré. But, moiré is just one of the possible artefacts. Another one is stippled hair.

I see little advantage with large pixels. Small pixels seem to carry a price, specially in medium format, but I am pretty sure manufacturing costs are pretty similar for small and large pixels.

Small pixels have less moiré, but very probable they also sharpen better and cause less interpolation artifacts.

So, yes, the sensor should outresolve the lens.

Best regards
Erik

There seems that there are preferences on both "sensor out resolving the lens" and "lens out resolving the sensor" sides… which side are you on and (most importantly) why? It seems that MF makers tend to favour the first for their higher end models…, are they right in doing so?

[Theodoros]

Hi,

Just as it happens I have updated a small article I have written on the issue:

http://echophoto.dnsalias.net/ekr/index.php/photoarticles/78-aliasing-and-supersampling-why-small-pixels-are-good

I don't think raw processors can handle moiré, they can reduce colour moiré. But, moiré is just one of the possible artefacts. Another one is stippled hair.

I see little advantage with large pixels. Small pixels seem to carry a price, specially in medium format, but I am pretty sure manufacturing costs are pretty similar for small and large pixels.

Small pixels have less moiré, but very probable they also sharpen better and cause less interpolation artifacts.

So, yes, the sensor should outresolve the lens.

Best regards
Erik

OTOH, one could argue that you get interpolation artefacts while capturing since lens resolutions spread on more than a pixel line or column and that they don't sharpen better for the same reason as before...

[ErikKaffehr]

Hi,

This may be worthwhile reading:
http://theonlinephotographer.typepad.com/the_online_photographer/2009/02/why-80-megapixels-just-wont-be-enough.html

Tim Parkin was looking at diffraction and found that a D800 at f/22 and found that it could be sharpened enough to surpass a Sony Alpha 900 at f/5.6. I am not sure I agree with him, but the article is here:
https://www.onlandscape.co.uk/2012/08/the-diffraction-limit-how-small-is-too-small/

Best regards
Erik

OTOH, one could argue that you get interpolation artefacts while capturing since lens resolutions spread on more than a pixel line or column and that they don't sharpen better for the same reason as before...

[BJL]

I can see arguments for having a sensor that somewhat outresolves the best lens in your system, at its optimal aperture and in center of field, so that the sensor will significantly out-resolve most of your lenses at most aperture choices, especially towards the edges of the frame.

The main one is economic: it is easier and less expensive to improve the resolution of a sensor in a given format than to improve the resolution of the lens system -- moreso for people who own and use multiple medium format lenses. So economics suggests that for a given expenditure, you can get more resolution improvement from sensor upgrades than from lens upgrades. In general, the cheapest component should somewhat outperform the most expensive component if you want to get the best results within a given budget.

Bear in mind that lens and sensor resolution combine in a "multiplicative" way (MTF values are multiplied), so starting with a lens and sensor of equal resolution, increasing the sensor resolution will improve the results that you get from the same lens, and increasing the lens resolution will improve the results that you get from the same sensor. But the sensor upgrade option is cheaper, especially when it improves the results from multiple lenses.

Another argument is simple physics: it is becoming feasible to produce sensors in formats 35mm and larger with resolution exceeding the fundamental optical limits of most lenses at most apertures of interest (due to diffraction for example) and when a lens is up against those limits, the only way to improve image resolution and detail is to push sensor resolution somewhat beyond the physical limits of the lens, in order to squeeze the most out of the lens. This is already the case when DOF needs require using an aperture ratio high enough that diffraction is the dominant limitation on the resolution given by the lens: about f/4 to f/5.6 I think, and a lot of MF work needs higher aperture ratios than that. Photographers seeking ever high resolution will more and more often be "diffraction limited" rather than "lens aberration limited", and then the last thing you want is to lose some of the lens' resolution by instead being "sensor limited".

EDIT: typos corrected.

[Theodoros]

Hi,

This may be worthwhile reading:
http://theonlinephotographer.typepad.com/the_online_photographer/2009/02/why-80-megapixels-just-wont-be-enough.html

Tim Parkin was looking at diffraction and found that a D800 at f/22 and found that it could be sharpened enough to surpass a Sony Alpha 900 at f/5.6. I am not sure I agree with him, but the article is here:
https://www.onlandscape.co.uk/2012/08/the-diffraction-limit-how-small-is-too-small/

Best regards
Erik

I know these articles Erik, I have some objections though… 1. Some confuse more sharpening (as a value in PS or other) with effective amount of sharpening and 2. If sensor out resolves the lens, one gets up to 3 (three) rows or columns of pixels that can be "contaminated"… if the opposite happens, it's no more than two… now for two columns (or rows) to equal three, a 6μm sensel has to be compared with a 9μm one… which I doubt will ever be the case for comparison. OTOH, you may find that some "fat pixel" sensors may show no pixelation at 200%, which surely means that upsampling artefacts will be less if one up-samples to print by linearising his printer's drivers… Finally, lets not forget the Bayer pattern existence and the result that the above may have on nine pixels rather than only four. Additionally, lets not forget that lens resolution is not evenly distributed (it's less towards borders and further more towards corners) across the sensor area as theories assume.

[Bart_van_der_Wolf]

Bear in mind that lens and sensor resolution combine in a "multiplicative" way (MTF values are multiplied), so starting with a lens and sensor of equal resolution, increasing the sensor resolution will improve the results that you get from the same lens, and increasing the lens rssolution will improve the results that you get from the same sensor. But the sensor upgrade option is cheaper, especially when it improves the results from multiple lenses.

Exactly. That's why it is not very meaningful to think in terms of 'outresolving'. The combined MTF result is dominated by the worst performing component, and the better the other component is, the closer to the optimum of the worst component one can get.

The sensor often sets a relatively hard limit to resolution by its sampling density (Nyquist limit), but lens corner performance may be a major limiting factor in other combinations.

Cheers,
Bart

[torger]

Outresolving the lens is best, in theory. No aliasing problems, you maximize the potential etc.

However small pixels and wide angular response does not combine too well, at least haven't seen that yet.

And higher resolution sensors push optical design into crazy space, ie the market does not satisfy with what they get from the lens but want more resolving power, and optical designs get more and more extreme, ie larger, heavier less robust and more expensive. I'm not particularly fond of Zeiss Otus type of lenses.

I like a balanced system the best, where resolution in the files are typical for what is needed, and 50-80 megapixels is a good number for "normal" large format style photography. Then don't make optics more complicated than it needs to be to resolve that.

[eronald]

Exactly. That's why it is not very meaningful to think in terms of 'outresolving'. The combined MTF result is dominated by the worst performing component, and the better the other component is, the closer to the optimum of the worst component one can get.

The sensor often sets a relatively hard limit to resolution by its sampling density (Nyquist limit), but lens corner performance may be a major limiting factor in other combinations.

Cheers,
Bart

One should factor in the type of degradation - as has been pointed out lens diffraction may be to a certain degree invertible.

Edmund

[Theodoros]

Never the less theory assumes that: A. Lens resolution is uniform across the sensor B. That photons enter the pixel wells directly (no angle involved). So, I guess the truth is somewhat complex since there may be image areas that both happen… i.e. there maybe a central area where lens out resolves the sensor and then the opposite happening towards the edges… I think Torger's opinion is nearer to the optimum, but (judging from my lenses on the Contax) I would say that with those lenses, 6.5-7.5μm size pixels are ideal. Lesser size wouldn't benefit wider apertures and the difference at optimum apertures would not make significant difference.
I wonder why makers don't offer a "FF" (54X40.5mm) sensor at around 40mp, the Rollei HY-6 was mostly offered with 7.2μm (what would make a 40mp FF sensor) pixel size backs and its performance was amazing… nothing to be jealous of than higher resolution backs even though the sensor was quite old.

[BJL]

One should factor in the type of degradation - as has been pointed out lens diffraction may be to a certain degree invertible. 

I would think that having more "samples per Airy disk" helps with that "diffraction deconvolution", so that is if anything an extra argument for over-sampling (having sensor MTF high than lens MTF).

[BJL]

However small pixels and wide angular response does not combine too well, at least haven't seen that yet.

That is a possible concern, but we must avoid worrying about cases where more, smaller pixels on a sensor of the same size makes things look worse "per pixel", when viewing images of higher pixel count at greater enlargement and so making defects more visible. The relevant question instead is about the quality "per image", or "per square mm of sensor", as with comparisons of same-sized prints, or with the same portion of the image displayed at the same size on-screen.

[ErikKaffehr]

However small pixels and wide angular response does not combine too well, at least haven't seen that yet.

Yes, that may be a real problem

And higher resolution sensors push optical design into crazy space, ie the market does not satisfy with what they get from the lens but want more resolving power, and optical designs get more and more extreme, ie larger, heavier less robust and more expensive. I'm not particularly fond of Zeiss Otus type of lenses.

The way I see it the Otus tries to achieve optimum performance at large apertures and minimise vignetting.  With a bit compromise much smaller lenses can be built, like the Sonnar 55/1.8 for the A7/A7r.

I like a balanced system the best, where resolution in the files are typical for what is needed, and 50-80 megapixels is a good number for "normal" large format style photography. Then don't make optics more complicated than it needs to be to resolve that.

From what I see from my experiments I would say 3-4 microns are maximum pixels size to avoid aliasing at medium apertures. With 6.8 microns on the P45+ it takes stopping down to f/16 to eliminate aliasing.

Best regards
Erik

[ErikKaffehr]

Hi,

This is 6.8 microns with an 80 (Planar 80/2.8 ) mm lens at f/8:



And this is 3.9 microns with another 80 mm lens (Vario Sonnar 16-80/3.-4.5) at 80 and f/8 and same distance (downscaled to match size):


Best regards
Erik

[Fine_Art]

That is a possible concern, but we must avoid worrying about cases where more, smaller pixels on a sensor of the same size makes things look worse "per pixel", when viewing images of higher pixel count at greater enlargement and so making defects more visible. The relevant question instead is about the quality "per image", or "per square mm of sensor", as with comparisons of same-sized prints, or with the same portion of the image displayed at the same size on-screen.

Finally we are starting to get to what matters. It is amazing to me that put a technical question in front of very bright people they will argue it in countless threads whether or not the technical issue matters one jot. Taking apart the science issue is the end in itself.

Summary: If the lens outresolves the sensor you risk moire. If the sensor outresolves the lens your images look like shit when pixel peeping at 100%. Nobody really cares either way.

You pick a long enough lens that the detail you want clear in the image is well rendered. End of story.

For a 180 degree landscape a 35 or 28 lens stitched. Don't waste your time with the massively distorted ultra-wides.
For wildlife, obeying park rules of keep 30 yards distant, you want the biggest longest lens you can afford/ carry.
For portraits a singe frame 85-135 lens

I assume everyone with a technical bend has done the spreadsheet of line pairs per image vs field of view. Assume a good lens reaches 3000LP/PH. Given a diffraction limited lens is limited by aperture what is the resolved angle of your lenses? Match your lenses to what you want in the image. You know everything you need. Go take pictures.