New article on overgaard.dk - "Leica 75mm Summilux-M f/1.4"

[pop]

Advertisement (gone after registration)

If the COC is much smaller than the apparent pixel size, then its size becomes irrelevant and can be considered to be a point. If the COC is much larger than the apparent pixel size, then the latter becomes irrelevant. If both are within a certain limit roughly the same size, you will observe much aliasing which may or may not produce the appearance of a sharper image. This much depends on the relative positions of the COC and the apparent pixel.

 

... I think.

[01af]

If the CoC matches the pixel pitch and you shoot at the computed hyperfocal distance, then we should see sharpness gains by focusing on infinity.

That's right—but Jaap believes otherwise.

 

 

I do not know if a CoC in the vicinity of the pixel pitch (and smaller) is practical.

Of course not. This is a purely academical discussion with absolutely no practical value.

 

 

However, no matter how I think of it, pixel/sensel size does matter.

Pixel size does matter for the question how big you can print your pictures before the indiviual pixels will become perceptible to the beholder. Otherwise, pixel size does not matter at all.

 

 

If the COC is much smaller than the apparent pixel size, then its size becomes irrelevant and can be considered to be a point.

In theory, no. No matter how small the circle of confusion—yet smaller circles will yield still sharper images ... in pure-math theory.

 

In real life, however, you're right. For several reasons, there's a limit from where further improvements of image sharpness are impossible. Where this limit is depends on several factors, including (but not limited to) lens quality and aperture used. In any case, the limit is at a size significantly smaller than the pixel pitch ... I'd guess for good lenses at not too small apertures it's somewhere between 1/5 and 1/20 the pixel pitch.

[CheshireCat]

Actually, the things under the Bayer filter should not be called "pixels" at all.

 

A pixel is an "atomic" picture element holding the complete information about one location within the image. The pit on the sensor and the corresponding digit in the raw image file holds the brightness value for one color channel only. The pixels in later steps of the processing chain are quite artificial, being reconstructed by some computing tricks.

 

Ok, let's call them sensel (sensor element).

 

The color of a pixel in the final image is a function f of a group of sensels (picture element) that occupies a finite area in the sensor. The diameter of the minimum circle containing all sensels used by the function f is the minimum full-color resolution of the sensor.

 

Applying the Nyquist-Shannon sampling theorem we can now compute the maximum acceptable CoC, hence the DoF.

[Berlinman]

Its an interesting discussio, but for me it sounds like there is a mixture betweeb two different things:

- one ist the classic concept of DOF. The limitation here ist with resolutions we have normaly with modern cameras and lenses only the human eye (with typical magnifications and view distances). in this concept Olaf us right: pixel size has no influence.

- the other one is the DOF with maximum sharpness, where you can crop ut to 1:1 an every pixel is sharp. Here the pixelsize is the limiting factor.

[CheshireCat]

In real life, however, you're right. For several reasons, there's a limit from where further improvements of image sharpness are impossible. Where this limit is depends on several factors, including (but not limited to) lens quality and aperture used. In any case, the limit is at a size significantly smaller than the pixel pitch ... I'd guess for good lenses at not too small apertures it's somewhere between 1/5 and 1/20 the pixel pitch.

 

These guys beg to differ:

 

Nyquist–Shannon sampling theorem - Wikipedia, the free encyclopedia

 

[pop]

In theory, no. No matter how small the circle of confusion—yet smaller circles will yield still sharper images ... in pure-math theory.

 

In real life, however ...

 

I think there's the rub.

 

Your argument separates the optical image from the digital one. This is, of course, for the discussion of the physics of the lens useful. With CoCs appreciable smaller than a sensor pit (or possibly even the apparent pixel) the CoCs become progressively more difficult to discern from points.

 

Also, we're arguing as if there was an infinitely thin plane where the image takes place and is converted into countable electrons. At the scale we're discussing I presume the depth of the sensor and the various regions within might become relevant as well.

[pop]

Advertisement (gone after registration)

These guys beg to differ:

 

Nyquist–Shannon sampling theorem - Wikipedia, the free encyclopedia

 

 

The sampling theorem is of limited use here. It defines - of course - well enough the size and location of point and disc shaped image elements that can be discerned and located. However, in photography we're not dealing with points but with edges. I think it quite apparent that the position and shape of an edge running through many apparent pixels can be determined by the viewer with greater accuracy than that of a single point.

[CheshireCat]

Its an interesting discussio, but for me it sounds like there is a mixture betweeb two different things:

- one ist the classic concept of DOF. The limitation here ist with resolutions we have normaly with modern cameras and lenses only the human eye (with typical magnifications and view distances). in this concept Olaf us right: pixel size has no influence.

- the other one is the DOF with maximum sharpness, where you can crop ut to 1:1 an every pixel is sharp. Here the pixelsize is the limiting factor.

 

If you think about it, the concept is exactly the same.

 

The ambiguity is in the term acceptable [circle of confusion].

It is obvious that an image is usually captured to be consumed by humans, hence the chosen viewing distance and individual retina resolution do subjectively define what is acceptable or not.

But these parameters have nothing to do with the exposure process, and the whole concept of "standard viewing condition" is ridiculous, especially since the advent of digital photography.

 

The only thing that matters is the amount of information stored in the medium.

Therefore [assuming a perfect lens] the DoF is a function of the sensor element size.

 

Everyone is then entitled to his own "viewing conditions"...

[CheshireCat]

The sampling theorem is of limited use here.

 

The sampling theorem is the key point here, and among other things, it tells us that it is useless (and possibly detrimental, due to aliasing) to have a lens that resolves subpixel resolutions.

 

However, in photography we're not dealing with points but with edges. I think it quite apparent that the position and shape of an edge running through many apparent pixels can be determined by the viewer with greater accuracy than that of a single point.

 

Your eyes are dealing with edges.

Digital photography only deals with points.

[jaapv]

 

In real life, however, you're right. For several reasons, there's a limit from where further improvements of image sharpness are impossible. Where this limit is depends on several factors, including (but not limited to) lens quality and aperture used. In any case, the limit is at a size significantly smaller than the pixel pitch ... I'd guess for good lenses at not too small apertures it's somewhere between 1/5 and 1/20 the pixel pitch.

So there IS a limit determined by pixel pitch. The only difference between us being the fraction of the pixel pitch. That is an argument (I guess we are talking about the edge effects of Bayer Filters, crosstalk between sensor elements etc. plus lens imperfections here) that I have no quarrel with.

[pop]

Your eyes are dealing with edges.

Digital photography only deals with points.

 

I usually make photograph for looking at. If my computer wants to have picture points, it can bloody well make its own pictures.

[CheshireCat]

I usually make photograph for looking at. If my computer wants to have picture points, it can bloody well make its own pictures.

 

I mean, your eyes and brain are inventing the edges, but in reality there is just a bunch of pixels.

In any case, I am not sure what you wanted to prove with this edges thing.

[Berlinman]

If you think about it, the concept is exactly the same.

 

The ambiguity is in the term acceptable [circle of confusion].

It is obvious that an image is usually captured to be consumed by humans, hence the chosen viewing distance and individual retina resolution do subjectively define what is acceptable or not.

But these parameters have nothing to do with the exposure process, and the whole concept of "standard viewing condition" is ridiculous, especially since the advent of digital photography.

 

The only thing that matters is the amount of information stored in the medium.

Therefore [assuming a perfect lens] the DoF is a function of the sensor element size.

 

Everyone is then entitled to his own "viewing conditions"...

 

 

you are an advocate if my second point. That is a view if DOF in the digital world - as You say, but it makes no real sense. The concept of DOF only makes sense if you see the whole process including the viewer and his eyes. With that you can calculate a DOF. If you change somerhing like the magnification if Your print etc. you will get an other DOF ! there us no absolute DIF out there.

[01af]

These guys beg to differ:

 

Nyquist–Shannon sampling theorem - Wikipedia, the free encyclopedia

Those guys are talking about a different topic.

 

 

The sampling theorem is of limited use here.

Exactly.

 

 

However, in photography we're not dealing with points but with edges.

In particular, we're dealing with making an image of a real-world scene. The sampling theorem doesn't apply here. It only applies to the sensor sampling the lens' output which is just part of the imaging chain. But the signal to capture for the camera to take a picture is not the lens' output but the real-world scenery, i. e. the lens' input. That's a fundamental difference.

[01af]

The sampling theorem is the key point here, and among other things, it tells us that it is useless (and possibly detrimental, due to aliasing) to have a lens that resolves subpixel resolutions.

Before explaining what the sampling theorem is telling us, you'd have to understand it first. In fact, all half-way decent lenses resolve sub-pixel detail—and you wouldn't want a lens that didn't.

[Jeff S]

I hope Thorsten is taking notes.

 

Jeff

[Jennifer]

***Yawn***. I've watching this thread descend into a argument between techies. Do you guys actually take photos or do you get off on analysing technical specifications and reading textbooks on advanced optics? And believe it or not some of you wonder why there aren't more women on this forum.

[pop]

***Yawn***. I've watching this thread descend into a argument between techies. Do you guys actually take photos or do you get off on analysing technical specifications and reading textbooks on advanced optics? And believe it or not some of you wonder why there aren't more women on this forum.

 

I don't think there are that many men who take an interest in those kind of things, either. However, I do think that the forum is large enough to hold a discussion or two which are of interest to just a few members.

 

And yes, I take a personal interest in technical things which contribute to the way my pictures would look should I ever undertake to take one.

 

[CheshireCat]

Before explaining what the sampling theorem is telling us, you'd have to understand it first. In fact, all half-way decent lenses resolve sub-pixel detail—and you wouldn't want a lens that didn't.

 

Sorry 01af, but you don't know what you are talking about. It would be more elegant to admit that sometimes you (as any human being) can be wrong. This will certainly help us digest your caustic posts a little better.

 

That said, I certainly want a lens that resolves 1/20th of [my current generation sensor] pixel, because I will be able to use it with newer sensors

[Advertisement]

Hallo,

Look here: Leica M

[CheshireCat]

Do you guys actually take photos or do you get off on analysing technical specifications and reading textbooks on advanced optics?

 

These two activities are not mutually exclusive.

In fact, if we call them A and B, and use a value of 1 for doing the activity and 0 for not doing the activity, then the exclusive-or function...

 

... just kidding