MTF Curves ~ Sensor Resolution

[k-hawinkler]

Advertisement (gone after registration)

But sensor resolution is not dependent on contrast - and lens resolution is,

 

True

 

making the two incomparable.

 

I still can compare and find it useful, say for 50% or higher values.

[k-hawinkler]

We can extract from Erwin Puts paper the following values for the APO-R 280/4:

 

Resolution: 250 lppm, Contrast: 50%, about 150 lppm can actually be recorded on film.

Resolution: 300 lppm, Contrast: 20%.

Resolution: 450 lppm, Contrast: ~0%.

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[Thomas Chen]

K.H.,

 

I found that there are 3 kinds of gamut for selection in the 5K i-Mac:

1. The best for this monitor

2. AdobeRGB 1998

3. sRGB

 

I switched among all the settings and the color renderings are totally different on the desk top.

 

Perhaps the setting 1 features what Mr. JaapV mentioned: 99% sRGB and 88%AdobeRGB coverage.

 

There is no telling how about selecting the setting of AdobeRGB.

 

Perhaps somebody who own the X-Rite checker may help to find out.

 

All the Best,

 

Thomas Chen

[mjh]

Since Leica (along with every other camera manufacturer I'm aware of) doesn't publish the specs of its low-pass system, which would need to include both the cut-off resolution and the slope of the transition, it is impossible to derive maximum resolution from the spec sheet.

But since there is no low-pass filter in Leica’s M models, what specs are there to publish?

Edited December 22, 2014 by mjh

[mjh]

photo 2 is a 100% crop of center, showing performance of the elmarit 28-2,8 II full open. Strong aliasing is visible, centered at 100 lppm, and from the fact that false detail is shown in the area between 100 and 200, we can roughly estimate lens resolution with good contrast up to 130/150.The lens is out resolving the sensor.

Note that the moiré is mostly colour moiré, implying that the lens is ‘outresolving’ the Bayer filter array, not necessarily the sensor as such. And it is the overly eager demosaicing algorithm creating those coulour artefacts – there is no moiré in the raw data. As I have argued here and in LFI before, the cause of moiré isn’t so much that the signal to be sampled contains frequencies exceeding the Nyquist limit, but that data about some point is taken to be representative for a larger area even when it clearly isn’t.

[jaapv]

We can extract from Erwin Puts paper the following values for the APO-R 280/4:

 

Resolution: 250 lppm, Contrast: 50%, about 150 lppm can actually be recorded on film.

Resolution: 300 lppm, Contrast: 20%.

Resolution: 450 lppm, Contrast: ~0%.

 

When the contrast is 0 how can there be linepairs at all?

 

I still can compare and find it useful, say for 50% or higher values

 

Out of interest: what usefulness do you find then?

Edited December 30, 2014 by jaapv

[JimKasson]

Advertisement (gone after registration)

I'm way late to this thread, but I've been working with the concepts involved for a while. The interactions between lens and sensor resolution in a Bayer-CFA sensor are complex.

 

Some of you may be interested in the graphs in this post, which, among other things, show 3D plots of f-stop, sensor resolution, and MTF50 for a simulated camera.

 

MTF vs pitch and f-stop for a simulated Otus | The Last Word

 

If you poke around you'll see many more posts on the same topic.

 

I'll be pleased to answer questions, and will consider bespoke simulation runs.

 

Jim

[pgk]

When the contrast is 0 how can there be linepairs at all?

Many many years ago when I was involved with MTF lens testing I seem to remember that as a rule of thumb 10% on the MTF graph is taken as output being indistinguishable from 0% - i.e. 10% is roughly the point at which we no longer have worthwhile imaging information. As I say, its a long time, but I'd query the relevance of any 'low' portion of the MTF graph to real world imagery.

 

[And of course, the whole endpoint of any test should be based on real world output, something which we are apt to ignore in our quest for the 'best' performance available - whatever that actually is].

Edited January 29, 2015 by pgk

[JimKasson]

There are (at least) two useful ways for looking at the relationship between sensor and lens resolution.

 

One is the concept of "System Q". I define and explore that here and in the subsequent posts:

 

Camera vs sensor resolution | The Last Word

 

Here is a more rigorous approach:

 

Optics InfoBase: Optics Express - Q Selection for an electro-optical earth imaging system: theoretical and experimental results

 

I find that precise application of System Q to Bayer-CFA cameras is imprecise, because of the ambiguity of the Nyquist frequency in those sensors.

 

A more heuristic, and, I believe, a more practical, approach lies in producing "quiver plots" of the 3D MTF50/pixel pitch/f-stop space, so that the arrows point in the direction of greatest improvement.

 

Sensel vs lens resolution | The Last Word

 

Here's an example, for a simulated Otus 55mm f/1.4 on a sensor with no AA filter:

 

 

The horizontal axis is f-stop, and the vertical axis is pitch in micrometers. If you consider that tthe M240 sensor resolutions is at the top part of the chart, you can see that, even at f/11, the steepest slope is in the direction of reducing pixel pitch. Looking at the 4.7 um pitch of the D810 and a7R (the finest today available in a FF camera), we see that statement to be true at apertures of f/11 and wider.

 

Jim

[JimKasson]

But since there is no low-pass filter in Leica’s M models, what specs are there to publish?

 

The MTF curves of a camera/lens system are highly dependent on the size and shape of the light-sensitive areas on the sensor.

 

Jim

[jmahto]

Bumping it up since there are many knowledgeable posts. Takes some time to digest.