Does the lens or sensor limit the resolution of an image?

[TonyW]

I did an experiment with my camera (Canon 5diii with 100mm F2.8 macro lens at F8) to see if I could determine what limited the resolution of the lens. I photographed a 30cm rule four times: the whole length, half the length, a quarter the length and an eighth.


056A3432 by tonyw36, on Flickr


056A3433 by tonyw36, on Flickr


056A3434 by tonyw36, on Flickr


056A3435 by tonyw36, on Flickr

If you look at the versions on Flickr, you see that the width of the images is the full 5760 pixels so that for the first picture of the full length of the rule there are 19 pixels to the millimetre. For the last image it is about 150 to the millimetre.

I zoomed in on each of the these images keeping one of the distance marks in the centre (I chose the 2 cm line) until the individual pixels were clearly visible. From the last image, we can see that the line is about 30 pixels wide, i.e. about 1/5 mm.

We can see on all the images that the change from white to black occurs over a something like 7 pixels, although most of the change occurs over about 4 pixels. (I think you have to see it to understand what happens but I have not been able to display an image which is only a few pixels wide here.)

My conclusion is that it is the lens which limits the resolution, not the pixel density.

Does anyone agree that this experiment makes sense? If so, it would be interesting for people with other cameras to do a similar test. Also, to test other lenses and different apertures.

[Colin Southern]

Are you applying capture sharpening? If not then your AA filter will be contributing to the softness as well.

[dje]

Sorry Tony but I think there's a fundamental flaw in your method viz to assume that the transitions from white to black on the ruler markings are somewhat sharper than the pixel pitch. Even if they were, I think you'd have to look at the individual pixel values in the raw file to see how they change. When you zoom the display closer than a 100% view, there is up-scaling going on in the display which will change the apparent sharpness.

Dave

[pnodrog]

You are making an assumption that the graduations on the ruler have been printed perfectly so for this experiment you are probably just analysing the ink spread on the ruler.

P.S. Sorry Dave I have just seen you have already spotted this flaw.

[AB26]

Tony,

I might be wrong. To my knowledge, what I have learned from spending hours reading camera stuff, the resolution is determined by the sensor only and sharpness is determined by the lens.
It is not resolution only that will render an image clear and sharp. The sharpness of the lens will determine how crisp an image will be. That is why some Leica cameras render images so sharp it will cut you when viewing on a good computer screen.

[pnodrog]

Tony,

I might be wrong. To my knowledge, what I have learned from spending hours reading camera stuff, the resolution is determined by the sensor only and sharpness is determined by the lens.
It is not resolution only that will render an image clear and sharp. The sharpness of the lens will determine how crisp an image will be. That is why some Leica cameras render images so sharp it will cut you when viewing on a good computer screen.

Andre you are correct. You have just reminded me of measurement theory I was forced to learn years ago which makes a very clear distinction between resolution and accuracy. In digital photography the analogy would probably be the sensor providing the resolution and the lens the accuracy.

[TonyW]

Some of this is getting too technical for me! But thanks for responding.

L. Paul and Dave, I did not make any assumptions about the gradations on the ruler and I did think about that question. That is why I photographed the ruler at different magnifications. At the highest magnification, about 1:1, the change from black to white occurs over a few (about 7) pixels, which is 7/150 mm, say 0.05mm as measured on the rule. This gives an upper bound to the distance over which the mark on the rule changes from white to black. At a lower magnification of say 1:4, the gradation on the image takes the same number of pixels which would be 0.2mm as measured on the rule. Hence my supposition that the result is due to the lens, not the pixels and not the fineness of the edge of the mark. I should do an experiment with an edge that is known to be sharp.

Andre, I am not sure what the difference is between sharpness and resolution. I guess I am really talking about sharpness. Whatever resolution is, I find it hard to believe that it is determined by the sensor if the sharpness is determined by the lens but perhaps this depends on what is understood by resolution. I don't think the question is to do with accuracy here.

Colin, you are really challenging my knowledge and understanding of signal theory. I did a little research with the help of Mr Google and it seems that there is an physical anti-aliasing process going on in the sensor even before the analogue signal is measured so that it seems reasonable that it could contribute to the softness.

The photographs were taken in raw mode which I thought meant that no further processing was done by the camera after the analogue to digital conversion. When I tried sharpening the images, the results were the same but again I don't know how sharpening algorithms work.

[Ady]

Hi Tony,

I'm sure I will be corrected if I'm incorrect, but as I understand the term 'resolution' I would have thought the answer to the question 'Does the lens or sensor limit the resolution of an image?' would be; whichever has the least capability of rendering the smallest separation or component parts distinguishable of (in this case) an image.

Assuming that the definition of resolution is a measure of the capability of a device or process in rendering distinguishable the smallest separation or component part, then the limiting factor will be whichever part of the device or process that is least capable.

So the lens OR the sensor (or any other part or process that has an impact on the ability of the camera to indicate or represent differences in output) may be responsible for limiting the resolution of the image.

CAVEAT: I have no formal/technical background in any of this so I may be talking pure nonsense based on poor understanding/interpretation from a laypersons perspective!

Cheers,
Ady

[chauncey]

It's a combination of both...http://www.luminous-landscape.com/tu...solution.shtml
http://www.canonrumors.com/forum/index.php?topic=3393.0
http://www.dpreview.com/forums/post/35963520
http://www.dxomark.com/index.php/Abo...ses-on-cameras

[pnodrog]

Adrian I seem to be agreeing with everyone tonight and partly because resolution is used in two different but related aspect in photography but ultimately the overall attainable "resolution" is determined by the weakest link. Up to at a guess about 12megapixels most quality lenses will out perform the sensor (assuming FF and ignoring anything other than resolving power ) but beyond that the performance of the lens becomes an increasingly significant limiting factor. By the time you get over 30megapixels the lens limitations really start to show. The trouble is resolving power is not the only issue regarding image quality.

A lot of testing has been done by http://www.dxomark.com/ and you may want to have a look at their comparisons of different lenses on different cameras or the same lens on different resolution cameras.

[Manfred M]

I think people are starting to "get it"; in any system, the weakest component in a system limits the overall system performance.

Using a bit of an analogy; let's look at the car. The Bugatti Veyron Sport Vitesse I believe is the fastest production car right now, with an average top speed of 408.84 km/h (254.04 mph). This of course means you have to have tires that can take the forces generated at that speed. Put on some tires that can only go up to 200 km/hr (120 mph) before they fail and you have the camera / lens analogy; a camera with a high quality sensor can be limited by the glass you mount on it (and how you use it).

The analogy can be extended; even with the right tires on the car, I probably would not be able to drive at that speed, while a professional race car (given the right conditions to do this with) would have no issues hitting top speed. A person with unsteady hands will not get the same image quality in a handheld shot as one taken using a heavy duty, weighted tripod.

[GrahamS]

The Nyquist limit is a function of the sensor and anti-aliasing filter assembly. Even if the lens can resolve beyond this limit, detail so resolved will not be reproduced in the final image. However, the contrast characteristics of a lens, as well as its colour performance and chromatic aberration, will affect the final image regardless of the nyquist limit of the sensor and anti-aliasing filter. The result is that an image produced by a low resolution but high contrast lens can appear to be sharper than the same image produced by a high resolution but low contrast lens.

[xpatUSA]

Unfortunately, the word "resolution" by itself means all things to all men (& ladies, to be PC).

And when it gets jumbled up with all kinds of generalities, confusion sometimes occurs.

A lens has a limit where it's contrast ratio drops to zero due to aberration at the big aperture end and mostly due to diffraction otherwise.

A sensor has a limit where it can not resolve items smaller than a pixel - and that is irrespective of AA filters and of micro-lenses and of pixel area and of Airy disks. Thus the determinant of sensor absolute resolution is simply the pixel pitch in the direction of interest.

As an example, my humble 12MP micro 4/3" has 'better' sensor absolute resolution than a D800e
.

[Glenn NK]

There is another issue at play here; the aperture at which the lens has the best resolution (sharpness).

This particular lens (Canon EF 100 f/2.8 macro), which I also have is not its sharpest at f/8. Diffraction is starting to have an effect (albeit it very slightly).

http://www.photozone.de/canon_eos_ff..._28_5d?start=1

Its centre sharpness is marginally better at f/4, but at f/5.6 the borders and corners are marginally better than at f/4. In spite of this, I usually use this lens at f/2.8 to f/4 as the real world differences are marginal, and the DOF limitations at larger apertures is more important (IOW creativity and composition in photography overrule tests).

Shortly after the Nikon D800 was announced, there were several discussions on other forums as to which lenses could "keep up" to the resolution of the sensor. I haven't heard this mentioned for about a year now, so I'm assuming that everyone finally got back to photography.

Glenn

[xpatUSA]

I should do an experiment with an edge that is known to be sharp

Sharp edges are quite hard to find, but those Aussie rulers are of "finest ENGINE DIVIDED quality" - at least that's what it says on my W&G model 1312

If you would print an almost black square on an almost white background about half the width on a A4 piece of paper, and then put it, tilted about 5 degs, on a wall maybe a meter or two away and take a shot. Publish your result unsharpened here and I will happy to tell you the edge spread, MTF, etc of your sensor and lens combination.

As has already been said - it's the sensor that limits the image resolution in the image plane at the sensor face.

[AB26]

Andre, I am not sure what the difference is between sharpness and resolution. I guess I am really talking about sharpness. Whatever resolution is, I find it hard to believe that it is determined by the sensor if the sharpness is determined by the lens but perhaps this depends on what is understood by resolution. I don't think the question is to do with accuracy here.

Tony,

I think of resolution and sharpness in terms of how an inkjet printer squirts dots on a page in patterns. If the print head squirts to much ink onto the page the ink flows before it dries and gives the dot a soft edge. If the print head squirts the right amount of ink on the page and it dries without flowing you get dots with sharp edges on the page. No matter how many dots you have per square inch, all the dots will be either sharp or soft depending on how good the print head is in squirting dots onto the paper.
Resolution is the amount of dots per square inch squirted onto the paper. The more dots the printer prints per square inch the higher the resolution and the less dots per square inch the lower the resolution.

A good lens will render sharp images even on a sensor with low resolution. A bad lens cannot render sharp images no matter how good the sensor. You can have an image that is way out of focus captured on a high resolution sensor and enlarge the image without any pixilation.
Resolution will be determined by the amount of photons a sensor can absorb.
The lens will determine how sharp the image can be rendered after “clean” photons was captured by the pixels on the sensor.

[revi]

There's two different aspects to a 'sharp' image:
- how finely resolved are the details, or 'what are the smallest details visible', and (resolution)
- how much edge contrast is there (acuity).
An image with somewhat less detail, but high edge contrast can appear sharper than an image with more detail, but less edge contrast.

The sensor imposes one absolute limit on the smallest details that will be visible: you need an edge to see a detail, so anything in the image smaller than a pixel won't be visible (more or less). The nature of light imposes another: you cannot focus light in an infinitely small area (a point) (see 'airy disc'). So those two determine the maximum resolution you can get.
The better the lens, the closer the resolution will be to the physical limit. But any optical system suffers from imperfections: chromatic and spherical aberration, astigmatism, internal reflection, ... In practice, these lead to less resolution, and a diminished contrast of the image, so less acuity. So the better the lens design, the less imperfections there will be, and the sharper the final image (both in resolution and in acuity).

The direction of the incoming light also plays a role: when you have direct light hitting the lens, you often get a softer image. That's due to reflections within the lens, leading to less contrast and less acuity in the image, while the resolution will be a lot less affected.

That's also why an unsharp mask appears to increase sharpness in the image: it increases edge contrast, and thus acuity. Try a high radius, like 10 or 15 pixels with a high value, like 100%, on an image of a dark subject on a light gray background, it'll give a nice bright halo around the subject. In normal use we try to keep those halos small enough to be invisible, but we still get the increased edge contrast. But this might obscure some of the finest detail in the image (only visible when pixel peeping at 100%, so absolutely irrelevant in normal situations).

A method like deconvolution sharpening can increase the actual resolution in the image (by removing part of the blurring from the anti-alias filter), but will not increase edge contrast, so the effect seems very limited. Also, used too aggressively, it will give unpleasant (and very visible) artefacts (due to the nature of the technique).

[xpatUSA]

Resolution will be determined by the amount of photons a sensor can absorb.

Sorry, Andre, can't let that pass. It is misleading and incorrect . .

. . unless there happens to be a credible reference that says so, of course.

[AB26]

Sorry, Andre, can't let that pass. It is misleading and incorrect . .

. . unless there happens to be a credible reference that says so, of course.

Hi Ted,
If my statement is indeed incorrect and misleading you should not let it pass. An explanation as to why you say it is incorrect and misleading will be appreciated. You should not challenge any statement without an explanation.

Looking forward to better understand your view of how high resolution is produced.

[AB26]

- how much edge contrast is there (acuity).

Remco,

Detail, resolution and sharpness are three different components in a rendered image. That is what I understand.
Am I perhaps missing the bus here?

Sharpening an image in PP definitely increases the file size, thus the resolution. How does that happened?