Will sharp ever be too sharp?

[Evan Cain]

Was just thinking about this today while out shooting..
The most recent super teles from Canon are pretty close to as good as you can get when it comes to their MTF charts, and you would have to be pretty good at spotting the difference between 500 mk1 vs mk2. Just a simple question, will there come a point when increases in sharpness from lenses, and a camera's ability to resolve detail will no longer be noticeable to the human eye. Making it virtually impossible to tell the difference from a lens/camera model's predecessor (assuming you are only measuring sharpness), without MTF charts?

How far away do you think Canon and Nikon are from this point, namely lenses.

Evan

[John Chardine]

Hi Evan- As I mentioned in a previous BPN post about sensor resolution, it has never been fruitful to predict limits to future science and technology development because when it is done it always turns out wrong when the time comes (like the stupid end of the world predictions!). We probably thought at one time that no one would ever need a sensor bigger than 8MP. In computer technology there was a famous quote attributed to Bill Gates that went something like "640k should be enough for anyone". Ten years ago would you have ever thought you needed a 4TB hard drive, and if so that it would be small enough to put in a briefcase?! The problem is that we take our demands and needs of today and do not do a good job at predicting what they will be in the future.

I think we are at a point now where sensor resolution is putting demands on modern lenses that sometimes cannot be met and I think it's fair to say that sensor resolution will continue to increase and lens quality will have to keep up. Nikon and Canon both make optics for the production integrated circuits on silicon wafers that far exceed the resolution of modern day lenses. So they are capable of giving us more on the lens side, although at what price to the consumer is a big question (can that level of precision be scaled up to a large lens at prices people could afford?).

One area that needs to develop hugely IMO is sensor technology- not resolution but the way in which images are formed. We are in a Bayer Array rut in my opinion with only a few niche cameras having different sensor technology (e.g., Sigma, Fuji). Not until you have seen the image quality from a non-Bayer array sensor, like Sigma's Foveon sensor for example (I own a camera with one) do you realise what modern lens and sensor technology are capable of (with a Bayer array the results can be approached or matched only by high-resolution sensors which lack an anti-aliasing filter, like Nikon's D800e).

I do think you are right though in suggesting that there are limits to what the human eye can resolve. However, I think our demand for ever more high resolution sensors and lenses that can meet their requirements will continue to rise as we demand more or more ability to zoom into portions of an image or make larger and larger prints to name just two future "wants"- I'm sure there are many others.

[arash_hazeghi]

Hi Evan,

when you compare two lenses that are similar in their characteristics (e.g. CA, coma, spherical aberration etc.) , as long as they out-resolve the image sensor i.e. the lens can render detail that is smaller than the sensor can sample there will be no visible difference in perceived sharpness. this is usually when MTF50 point is above the Nyquist limit for an image sensor. The Canon MK1 and MK2 supertelephoto lenses (without extenders) out resolve all current Canon sensors so you will not see a difference in sharpness.

here is more info:

http://www.imatest.com/docs/sharpness/

As for limits of resolution it is goverened by laws of diffraction assuming everything else is perfect which is not the case usually. If you make the pixels too small beyond the diffraction limit they will no longer resolve more detail for given optics. This limit is determined by the numerical aperture of the system.

http://en.wikipedia.org/wiki/Diffraction-limited_system

here is useful calculator that shows how close a camera at a given aperture is to diffraction limit http://www.cambridgeincolour.com/tut...hotography.htm

usually imperfections in the optics kicks in before this limit, but for super-telephoto lenses that are almost perfect this limit is closer to reality.

[Tom Graham]

"...camera's ability to resolve detail will no longer be noticeable to the human eye..."
We are already there.
When you view an 8x10 inch print from 10 feet away, you can't tell whether the camera was a 1 mega pixel or 10 mega pixel.
Tom

[John Chardine]

"...camera's ability to resolve detail will no longer be noticeable to the human eye..."
We are already there.
When you view an 8x10 inch print from 10 feet away, you can't tell whether the camera was a 1 mega pixel or 10 mega pixel.
Tom

Viewing distance is an important factor, you are right Tom - the "billboard effect". But I think it's best to consider what you might call "normal" viewing distances. Ten feet for an 8"x10" is too far. Are we "there" now?

[Tom Graham]

Yeah, I know.
Sad that Roger C. is no longer moderator here BPN. He is a world reknown scientist in image processing. Especially astronomical. These type of camera sensor/pixel questions are no challenge for him!!!
Sooo, you might visit his web site, he discusses and tests much about camera resolution from theoretical and practical perspectives.
http://clarkvision.com/ and super lion photo on home page now (photo changes every so often).
Tom

[Evan Cain]

Thanks everyone for your responses. Looks like I have some reading to catch up on :)

[Tom Graham]

riddle me this -
".... the higher resolution ones are so high-resolution that they almost have to retouch the sharpness out of them. I couldn’t tell you which one we’re using, but we’re not using the highest-end ones because people complained that they were too clear'. 'Film was never this sharp. It’s sharper than real life. You shouldn’t be able to read a hair inside the tear duct of someone’s eye. On one of those high-end backs, though, you can almost read what someone is thinking. It’s kind of terrifying'."

from - http://www.dpreview.com/news/2013/08...er-interviewed

Tom

[John Chardine]

Interesting article (never seen the show). I think for some applications such as people portraiture, resolution and sharpness isn't necessarily what you want. However, I would rather start with a high-res, sharp image and process the softness into it rather than start out with one out of the camera.

[John Chardine]

Just saw this by Michael Reichmann.

http://www.luminous-landscape.com/es...ick-wall.shtml

Essentially repeats what has been said here but the last line in the article caught my eye:

"If someone produces a 35 mm full frame camera with 100 Megapixels, beware. Given the limitations of the wavelength of light, no lens can live up to that resolution. I suggest that the Megapixel race is over. Marketing people may produce one of these cameras, but no physics expert will buy one."

I note with some interest that although Nikon has ventured into the realms of 36MP FF and 24MP crop (1.5), Canon has steadfastly refused to do this thus far. Perhaps they can't but maybe they just won't, at least right now.

[Roger Clark]

Just saw this by Michael Reichmann.

http://www.luminous-landscape.com/es...ick-wall.shtml

Essentially repeats what has been said here but the last line in the article caught my eye:

"If someone produces a 35 mm full frame camera with 100 Megapixels, beware. Given the limitations of the wavelength of light, no lens can live up to that resolution. I suggest that the Megapixel race is over. Marketing people may produce one of these cameras, but no physics expert will buy one."

I note with some interest that although Nikon has ventured into the realms of 36MP FF and 24MP crop (1.5), Canon has steadfastly refused to do this thus far. Perhaps they can't but maybe they just won't, at least right now.

Let's try some simple physics.

OK, you have a nice telephoto and 22 megapixel full frame camera. You add a 2x TC and get great images, right? The TC did not improve the resolution of the lens, it only magnified the image at best (and probably degraded image quality a little). But you still get a lot more detail on the subject. The TC is functionally equivalent to using smaller pixels for the same pixels on subject (without degradation of image quality from the TC optics, nor the slowing of AF due to slower f/ratio). Without the TC, one would need to halve the linear pixel size for the same pixels on subject. Thus, the megapixel count would go up 4x, or 88 megapixels.

Now try it on a 1D Mark IV: a supertelephoto and 2x TC and it produces very nice results (many such images are posted here on BPN). Without the 2x TC, one would again need to halve the pixel size for the same pixels on subject, and then scale to full frame (1.3*1.3) and we would have 16 * 4 * 1.3*1.3 = 108 megapixels.

So no, sensors are not outresolving some current lenses. Whether one needs that number of megapixels is up to the use of the image.

Roger

[John Chardine]

Interesting Roger. So here's a "thought experiment". The Canon S100 "point and shoot" has approx. 1.9 micron pixels. Scaled to FF this would be a 250MP sensor! Assuming a tiny sensor like is in the S100 could be scaled to FF, would the images out of the camera be any good? BTW, the diffraction calc. at Cambridge in Colour suggests this sensor would not be diffraction-limited even at f22.

[Roger Clark]

Interesting Roger. So here's a "thought experiment". The Canon S100 "point and shoot" has approx. 1.9 micron pixels. Scaled to FF this would be a 250MP sensor! Assuming a tiny sensor like is in the S100 could be scaled to FF, would the images out of the camera be any good? BTW, the diffraction calc. at Cambridge in Colour suggests this sensor would not be diffraction-limited even at f22.

Hi John,
I do not see where the site says it is not diffraction limited. Are you referring to here:
http://www.cambridgeincolour.com/tut...hotography.htm
Perhaps your circle of confusion is set high. Go to advanced and check the box for setting the circle of confusion based on pixels. At f/22, the Dawes limit (0% MTF) for green light is 91 line pairs per mm, or pixel pitch = 5.5 microns at Nyquist sampling.

Small pixels have another issue besides diffraction, and that is the absorption length in silicon. The absorption length (1/e or 1/2.72, or 63% absorbed) is 5 microns for red light and over 3 microns for green light. So pushing much below that in pixel size has other side effects besides diffraction.

But contrary to popular belief, small pixels do not cause noisy images. The lens deliver the light. Smaller pixels simply chop the light into smaller pieces. For the same size display (e.g. print or monitor), usually the image made with smaller pixels will appear better if it is resized well (cubic spline is not the way to go in my opinion). Given a high megapixel image, one can choose the output resolution, trading noise and spatial detail. But with good pixel management in post production, better images can be made using smaller pixels. Imagine choosing if you want a 1.4x or 2x TC in post production! I show examples of pixel management in my series on exposure:
http://www.clarkvision.com/articles/index.html#exposure
Specifically, Figure 1 in part 3 shows this very case: the example shows an image from a 7D with its (undeserved) reputation for high noise compared to an image from a 5DII.

So I think 250 megapixels is pushing beyond limits due to both diffraction, and absorption length in silicon, but perhaps a 100 megapixels would be sweet if the frame rate could be kept high. For now I would be thrilled with 40-ish.

Roger

[John Chardine]

I had the calculator set up incorrectly Roger. I neglected to click the "Set circle of confusion* based on pixels?" option. When I do, the diffraction limit is f4, which makes sense.