Do sensors "outresolve" lenses?

[Phinius]

Having read the latest article by Ruben Osuna and Efrain Garcia I have the impression that the benefit of a high-megapixel (20+) 35mm-type cameras may not be worth the cost absent some particular need. This would seem especially true if we use the rule of thumb that to double resolution we need to quadruple megapixels.

If I understand the article, a big "if," an increase in f-stop beyond f5.6-8.0 actually reduces the usable number of megapixels from the sensor's maximum. If this is true, then I would be spending thousands of dollars to gain megapixels only at less than those f-stops. Furthermore, this increase would only be marginal because a 25 megapixel sensor does not have twice the resolution of a 12-megapixel sensor, but only would have a 25% increase below f-5.6 or so, and this would seem to come with increased noise all things being equal.

Even more confusing, if I were trying to decide whether to purchase either a 35mm-type camera or a medium format camera, whatever I gain in resolution in a medium format camera I may loose in lens quality given the physics of larger lenses.

Bottom line, why not just buy a state-of-the-art 12 megapixel 35mm-type camera?

Anyone disagree?

rmj@violetcrownphotographs.com

[Ray]

Bottom line, why not just buy a state-of-the-art 12 megapixel 35mm-type camera?

Anyone disagree?
[a href=\"index.php?act=findpost&pid=200983\"][{POST_SNAPBACK}][/a]

I'd go along with that but mainly because I haven't got many state-of-the-art lenses which are noticeably (or significantly) sharper at F5.6 than at F8.

Nevertheless, one can get a fair idea of the resolution differences between a 12mp FF camera like the 5D and a 25mp camera such as the upcoming Sony FF by comparing the 12mp 5D with the 10mp 40D, using the same lens at the same aperture, then cropping the 5D shot to the same FoV as the 40D shot. (The pixel density of the 40D is equivalent to that of a 26mp FF sensor.)

I get the impression that both the 20D and 40D, with a fairly average lens like the Canon 100-400 at 400mm and F8, produce a marginal but noticeable improvement in resolved detail compared with the 5D. If this lens were sharper at F5.6 than at F8 (instead of the reverse), then I imagine the improved detail would be very worthwhile.

It is axiomatic that sensors with greater pixel density need sharper lenses for those sensors to deliver the greater detail they are potentially capable of. Some P&S cameras seem to deliver sharpest results at F2.8, but most 35mm lenses seem to be sharpest somewhere between F5.6 and F8, or equally sharp at both F5.6 and F8.

[SeanFS]

I'd go along with that but mainly because I haven't got many state-of-the-art lenses which are noticeably (or significantly) sharper at F5.6 than at F8.

Nevertheless, one can get a fair idea of the resolution differences between a 12mp FF camera like the 5D and a 25mp camera such as the upcoming Sony FF by comparing the 12mp 5D with the 10mp 40D, using the same lens at the same aperture, then cropping the 5D shot to the same FoV as the 40D shot. (The pixel density of the 40D is equivalent to that of a 26mp FF sensor.)

I get the impression that both the 20D and 40D, with a fairly average lens like the Canon 100-400 at 400mm and F8, produce a marginal but noticeable improvement in resolved detail compared with the 5D. If this lens were sharper at F5.6 than at F8 (instead of the reverse), then I imagine the improved detail would be very worthwhile.

It is axiomatic that sensors with greater pixel density need sharper lenses for those sensors to deliver the greater detail they are potentially capable of. Some P&S cameras seem to deliver sharpest results at F2.8, but most 35mm lenses seem to be sharpest somewhere between F5.6 and F8, or equally sharp at both F5.6 and F8.
[a href=\"index.php?act=findpost&pid=201020\"][{POST_SNAPBACK}][/a]

I thought with the 14nx Kodak I had at 14mp and no AA filter it seemed to get all possible detail with a good quality lens in a 35mm FF system, but meant having to spend extra time removing noise from some detail in some situations. The 1ds2 is roughly equivalent in resolution ( but a far better built camera) Both cameras slightly edge out the 5d , but only by a small amount. The 1ds3 AA filter evens up the megapixel equation as far as fine detail,  but removes much of the moire and other noise at the same time. In my tests both cameras seem to do their best at f8 - f11 with a standard 50mm lens
The 1ds3 however, Is not a great jump. Images that seem tack sharp wit the 1ds2 at f11 seems a little soft on the 1ds3 at f11 - same amount of detail , just softer at 100 percent view, no matter how I sharpen. There is an improvement at F5.6 - f8 so its clear diffraction becomes an issue earlier with the larger pixel density, but its quite small. Personally I think the 1ds2 is pushing the quality of most 35mm lenses ( all of my Canon and Nikon ones , anyway) not quite to the limit and the 1ds3 just over in most cases. As far as optical design goes there are clearly challenges to be met using a sensor with more megapixels than even the 16 of the 1ds2 and even the old 14nx.

[smthopr]

Even more confusing, if I were trying to decide whether to purchase either a 35mm-type camera or a medium format camera, whatever I gain in resolution in a medium format camera I may loose in lens quality given the physics of larger lenses.

Bottom line, why not just buy a state-of-the-art 12 megapixel 35mm-type camera?

Anyone disagree?

I read the article but could not really follow the technicals closely.  That said, from personal experience, I find it hard to believe that a medium format digital camera can not out resolve a 35mm sized digital camera using the same angle of view. I just don't believe it.

[Ray]

I read the article but could not really follow the technicals closely.  That said, from personal experience, I find it hard to believe that a medium format digital camera can not out resolve a 35mm sized digital camera using the same angle of view. I just don't believe it.
[a href=\"index.php?act=findpost&pid=201047\"][{POST_SNAPBACK}][/a]

May be, but the topic of the thread is not about one format outresolving another, but the sensor outresolving the lens.

The concept of the sensor outresolving the lens means that more pixels on the sensor produces insignificant increases in resolution using the same lens. This can apply to all formats.

MFDBs usually have an inherent resolution advantage as a result of not having an AA filter, even though the number of pixels may be the same. The claimed softness of 1Ds3 images is apparently due to its AA filter.

[Geoff Wittig]

Personally I think the 1ds2 is pushing the quality of most 35mm lenses ( all of my Canon and Nikon ones , anyway) not quite to the limit and the 1ds3 just over in most cases. As far as optical design goes there are clearly challenges to be met using a sensor with more megapixels than even the 16 of the 1ds2 and even the old 14nx.
[a href=\"index.php?act=findpost&pid=201024\"][{POST_SNAPBACK}][/a]

For what it's worth, I had an interesting exchange with Ctein (technical photo guru and all around genius) at another site. He largely agreed with the article by Ruben Osuna and Efrain Garcia, noting that "these guys are GOOD". He notes that there is a hard, unbreachable physical resolution limit at about 80 megapixels for a full frame 35 mm sized sensor, but practical limits with real world lenses obviously kick in a lot earlier. My experience certainly matches Osuna & Garcia's numbers; using a very sharp lens with perfect technique and judicious post-capture sharpening, there is a barely perceptible improvement going from the 16 mp Eos-1Ds II to the 21 mp 1Ds III at very large print sizes. The improvements in high ISO noise control and all around usability are more noticeable than the nominal resolution increase, which appears to be bumping up against practical physical limits.

[sojournerphoto]

For what it's worth, I had an interesting exchange with Ctein (technical photo guru and all around genius) at another site. He largely agreed with the article by Ruben Osuna and Efrain Garcia, noting that "these guys are GOOD". He notes that there is a hard, unbreachable physical resolution limit at about 80 megapixels for a full frame 35 mm sized sensor, but practical limits with real world lenses obviously kick in a lot earlier. My experience certainly matches Osuna & Garcia's numbers; using a very sharp lens with perfect technique and judicious post-capture sharpening, there is a barely perceptible improvement going from the 16 mp Eos-1Ds II to the 21 mp 1Ds III at very large print sizes. The improvements in high ISO noise control and all around usability are more noticeable than the nominal resolution increase, which appears to be bumping up against practical physical limits.
[a href=\"index.php?act=findpost&pid=201100\"][{POST_SNAPBACK}][/a]

I've run some limited practical tests and the 1Ds3 provides clearly better higher resolving prints than a 5D - if using a good lens (100 macro) at apertures around f4 to 5.6. Diffraction starts to become visible at 100% at about f8 initially, so I'm not surprised that at f11 there is no more detail than frmo a 1Ds2. Diffraction is just about becoming visible on a 5D at f11. Given these criteria I'd say that the 1Ds3 give slightly better prints at 24 by 16 and substantially better at 48 by 32 than the 5D.

I showed prints to my wife and she picked the 1Ds3 prints as preferable in each case, but that's not to say the 5D doesn't make nice pictures!!

Practically, I suspect that there's not much more to come from 35mm format (or maybe larger formats) apart from stitching and accepting limited dof. My sense last year was that the sensible limit for 35mm was around 35Mp - although perhaps a foveon type solution would be benefitical. I'm not sure.

Finally, doubling the total number of pixels increases resolution by 41%, not 25%. Perhaps equally improtant though is that objects have area and this really does benefit from the actual increase in the number of pixels.

Mike

[Ray]

Finally, doubling the total number of pixels increases resolution by 41%, not 25%. Perhaps equally improtant though is that objects have area and this really does benefit from the actual increase in the number of pixels.

Mike
[a href=\"index.php?act=findpost&pid=201181\"][{POST_SNAPBACK}][/a]

It's important to bear in mind that doubling the pixel count increases the potential resolution of the sensor by 41%. In practice, that 41% increase will never be achieved without a corresponding (and very significant) increase in lens resolution, so the figure of 25% is a more realistic expectation.

[Paul Kay]

I'd hazard a guess (actually an educated guess following discussions with a lens designer) and suggest that whilst sensors will no doubt increase in their MPixel count, QC/cost issues will limit lens's abilities to keep up with these come what may. A quick scan of most photo fora will quickly reveal numerous complaints about lenses needing adjustment and the 1DSMkIII has adjustment built in to compensate for focus adjustment (so I read). I'd echo the Canon 100mm macro as being a very good lens, capable of outstanding results but many others aren't and improving them might well push them into hi-fi territory as the law of diminishing returns kicks in. (I use a 1DS which is perfectly capable of revealing lens flaws, and an M8 which does have stunningly good and correspondingly expensive lenses).

We do have some excessively good cameras with us NOW. Exploiting them to their full potential is not always as easy as its made out (IMHO) already!

[John Clifford]

I read the article, and the gist of it seems to be that anything beyond 15 MP is wasted in APS-C-sized Bayer sensors.

(I'm going to use 'pixel' here to indicate photodiode size... incorrect, but common terminology.)

The article also says that a Bayer sensor resolves at about 70% of theoretical (what you'd expect based upon # of pixels per mm), so the effective pixel size is really 1.6x the actual pixel size -- any detail that will not cover this much or more will not be correctly detected by the sensor. My experience supports the article's conclusion; while I have more pixels in an image from the K20D, I don't see more resolved detail, just 'blur'.

Interestingly enough, my K20D has a 5 micro-meter pixel diameter, while my SD14 has a 7.8 micro-meter pixel diameter, i.e., 1.6x the K20D! This means that the effective resolution of both sensors is identical -- on actual detail resolution both sensors are equal. However, the SD14 does have one advantage; it is diffraction-limited at f/16 while the K20D is diffraction-limited at f/11.

I would love to see Foveon bring out a 10 MP FF sensor.

[Ray]

I would love to see Foveon bring out a 10 MP FF sensor.

Yes. You might get close to the resolution of a 1Ds3, but I fear the noise level at high ISO would be very unimpressive. The Nikon D3 and D700 have set the standard for low noise. Future full frame sensors will have more pixels than the D700. However, if total image noise is worse, it will be considered a disadvantage and a negative.

Outresolving the lens is a good thing, provided total image noise is not compromised.

[Hening Bettermann]

Studying the article by Rubén Osuna and Efraín García, I have a question. The text preceding table 3 says that "Typical DSRL sensors have pixels in the range from 5 to 6.4 microns." The table (row 2 ) has values w= 0,4 - 0,55 - 0,7 microns (µ). Thinking that w was pixel width, I had exspected 4 - 5,5 - 7 µ. Where is this factor 10 coming from?
Thanks for clarification!

[Gary Yeowell]

Once i went up from my 5D to the 1DS3 i had to replace most of my lenses, which seemed fine on the 5D but lacking on the 1DS3!!!

I kept my great copy of the 35f1.4L and got rid of all my other 'L' glass, replaced them with 21 & 28 Contax Zeiss, 50 ZF Zeiss, Nikon 105f2.5AI. These lenses at apertures up to f5.8-F8 show what the 1DS3 is capable of, Canon zooms will not.

[Ray]

Once i went up from my 5D to the 1DS3 i had to replace most of my lenses, which seemed fine on the 5D but lacking on the 1DS3!!!

Surely you didn't have to. You chose to. Any sensor will record greater accutance, sharper results and slightly more detail in the low contrast areas of the scene, if used with a better quality lens. I never used Canon's first DSLR, the 3mp D30, but I recall reports from those who did use that camera, that results were improved when using a good quality prime instead of a lower quality zoom.

Basically, if you improve the lens without improving the sensor, you should get at least marginally better results. If you improve the sensor without improving the lens, you should also get at least marginally better results. If you improve both sensor and lens, then you can of course expect a greater-than-marginal improvement.

[Gary Yeowell]

The point being Ray that in terms of picture quality alone, IMO there's no point in going from a 5D/1DS2 to the 1DS3 if you have already hit the resolution limit of the lenses with current sensors. In my case, yes i HAD to, once i'd seen what a pile of shit my hand picked version of the 16-35mk2 was compared to the Zeiss 21/28 on a 1DS3 there was no going back. Also i was coming from using a Phase back on a Hasselblad and  the 1DS3 replaces both the 5D and Phase in one go as far as i'm concerned.

Of course there are many other reasons Ray, to own the 1DS3, biggest of all being live view and sensor cleaning as well as functionality, so many people will not feel the need to change lenses, and that's fair enough.

Gary.

[Ray]

The point being Ray that in terms of picture quality alone, IMO there's no point in going from a 5D/1DS2 to the 1DS3 if you have already hit the resolution limit of the lenses with current sensors. In my case, yes i HAD to, once i'd seen what a pile of shit my hand picked version of the 16-35mk2 was compared to the Zeiss 21/28 on a 1DS3 there was no going back. Also i was coming from using a Phase back on a Hasselblad and  the 1DS3 replaces both the 5D and Phase in one go as far as i'm concerned.

Of course there are many other reasons Ray, to own the 1DS3, biggest of all being live view and sensor cleaning as well as functionality, so many people will not feel the need to change lenses, and that's fair enough.

Gary.

That's simply not true, Gary, hitting the resolution limit of lenses with current sensors, unless you are talking about realy bad performance. I have personally gone from a 5D to a simulated 39mp full frame format (cropped to an APS-C format in the form of the 50D) and found that there is at least some marginally greater detail at F22. As you probably know, all lenses can be considered equal at F22.

You can never hit the resolution limit of a good lens used at its optimum aperture. As lens resolution gets finer, the contrast just gets fainter and ultimately sensor noise and lack of sensitivity becomes the limiting factor as pixel density increases.

The Zeiss wide angle lenses were also much appreciated by owners of the 11mp 1Ds. You don't need a 1Ds3 in order to appreciate the high accutance, low CA qualities of Zeiss glass.