Sharpness and resolution, one subject or two ? (1 Viewer)

Ron,

Thanks for sharing some of your data with us.

I'm a little confused by that Monarch 8x36 MTF result. Is that the right plot you attached? You said it had a resolution of 7 arcseconds but when I apply your formula to convert cycles/degree to arcseconds using 8x as the magnification I get something wildly out. Is there some other magnification in the optical train that should be included as well?

I'm supposing that was done using your Nikon Coolpix 5700 which has the pixel dimensions shown, and 2560x1920 pixel dimensions. The MTF50 figure is 10.4c/d but a value of 1012 Line Widths per Picture Height value is shown. That would be 1012/2 = 506 Line Pairs/ Picture Height. That works out a 1920/506 = 3.8 pixels/cycle or (3.8 x 10.4)/14.3 = 2.8 pixels/cycle for the MTF5 value if I've got it right.

Apologies if I've misunderstood something but if my numbers are right doesn't that rather look rather like the plot is the MTF for the Camera not the binocular?

Regards,

David

David

I tried to access the actual data file but could not find it quickly so did not search too long.

I believe the plot is correct. This was done 6/2007 and made a lot of changes since then but notice the diffraction limited line crossing the right border at 40% contrast. On a perfect MTF 40% occurs about half way along the x axis so this would extend to 60 c/d and 7" would be about 64 c/d.

The biggest problem I had with that software program was that it would only generate a max of 2x Nyquist which was very limiting for poor optics and it showed the first 5% contrast point it came across where the sample pixels would cross the 5% contrast limit many times.

While the plot would scale to cyc/deg the actual data dump to a csv file was in cyc/pixel so I had to do the actual plots at that time in Excel to do the conversion to extend past 2x Nyquist. Also the target was not a standard picture so picture height did not apply. The target in those days had an ROI of about 30x100 pixels so picture height varied.

Extended out the plot would look a lot like those for the 2 wave SA or 2 wave astigmatism in the response to Kimmo above.

I am mostly going from memory here and if I find any difference when I get a new CPU fan I will get back to you.

PS: On that camera 1 pixel is about 10"/8X or 1.25".

Ron,

I did have a go at the ImageJ with the slanted edge add on, but the lenses I have for the machine vision cameras weren't up to job so didn't get very far. I presume this in was Inmatest software. I've not used it but doesn't Page Height in their terminology simply mean number short axis sensor dimension, not the test image. So doesn't that mean that you've simply run out of pixel resolution at 15 cycles/per degree. 60 c/d would work out around 0.66 pixels for every cycle. I don't know what lumps and bumps are doing there at higher frequency but is it possible there is a bit of aliasing at 2 and 1 pixels/cycle?

David

Surveyor said:

No Bill. I have a cheap refraction program, nothing close to that kind of power,

Does it have some pertinent functions that could help?

Click to expand...

Perspective! I have Zemax-EE and two hard keys. Yet, it hasn't been turned on in over ten years. Thus, who is better off? |8|| I thought if you had it running, you could show some graphics of some of the stuff folks talk about.

Bill

WJC said:

Perspective! I have Zemax-EE and two hard keys. Yet, it hasn't been turned on in over ten years. Thus, who is better off? |8|| I thought if you had it running, you could show some graphics of some of the stuff folks talk about.

Bill

Click to expand...

No idea of which of us is better off. Last time I looked at it the price was almost $4k for a single seat of the bottom version. A bit out of my range to use an hour or so every couple of years.

Refresh my memory, isn't EE the older version with ZPA that was about 12k per seat and had all the lighting engineering functions in it?

typo said:

Ron,

I did have a go at the ImageJ with the slanted edge add on, but the lenses I have for the machine vision cameras weren't up to job so didn't get very far. I presume this in was Inmatest software. I've not used it but doesn't Page Height in their terminology simply mean number short axis sensor dimension, not the test image. So doesn't that mean that you've simply run out of pixel resolution at 15 cycles/per degree. 60 c/d would work out around 0.66 pixels for every cycle. I don't know what lumps and bumps are doing there at higher frequency but is it possible there is a bit of aliasing at 2 and 1 pixels/cycle?

David

Click to expand...

David, this was about 8 years ago and my memory has gone elsewhere, have not found it yet. Thinking about this, about that time frame I had a couple of other cameras that had the same pixel size or close to it but only an 8x6 mm sensor instead of the 6.6 mm height. I think it may have been an Olympus of some type. I have never used picture height for any sort of use so I am unfamiliar with it.

The old Imatest I had, I think it just had a check box for anti-aliasing. The current one I am using has 0, .2 or .4x anti-aliasing for the custom portion but I think the ISO function is preset but I don't remember if it is off or 2x.

There may be some aliasing going on but I doubt if it is much. Using the ISO default and checking from one program to the next with the same images I got very much the same results.

I found an image of a reduction of the same image data in the folder with the one posted but I think I may have done this with a demo version of something I bought later. I have no notes on it but it looks very similar with the exception of showing all the channels.

The way SFR works is the slant is set up along a chosen axis so that the way it is sampled you sample 10 pixels in one direction for 1 pixal perpendicular. This gives a 0.1 pixel (assuming 5°) level detection for each pixal along the other axis. I have looked the equations up at one point and can honestly say I have no idea of what they are doing, a lot like the Roddier reduction of star tests.

Thanks Ron for the additional data.

As you say it appears to pretty much to match the first plot with almost identical MTF5 value at 2.85c/p so presumably the pixel size was the same or very similar. It again seems pixel limited so isn't it camera resolution? The lumps and bumps are shown a bit further out but no signs of a harmonic this time. Don't know what that means.

Cheers,

David

Surveyor said:

No idea of which of us is better off. Last time I looked at it the price was almost $4k for a single seat of the bottom version. A bit out of my range to use an hour or so every couple of years.

Refresh my memory, isn't EE the older version with ZPA that was about 12k per seat and had all the lighting engineering functions in it?

Click to expand...

Yeah, its the big boy. But, it's not the software; it's the mind behind it. Like all my former guitars, I had the goods but NOT the talent. The average Zemax is over 5K. I don't know about the E package.

When I bought Zemax, it was yet to be 1.0, came on A 3.5-inch floppy, and was purchased while Ken was on his way to his first job out of UA. (read: a lot cheaper). But, Ken sold it and is not in the driver's seat anymore.

Boy, he sure could write tight code. I may be just a screw turner, but I could recognize that.

Cheers,

Wannabe Bill

typo said:

Thanks Ron for the additional data.

As you say it appears to pretty much to match the first plot with almost identical MTF5 value at 2.85c/p so presumably the pixel size was the same or very similar. It again seems pixel limited so isn't it camera resolution? The lumps and bumps are shown a bit further out but no signs of a harmonic this time. Don't know what that means.

Cheers,

David

Click to expand...

David, I am not sure I understand what you are asking but this may help. The only thing that matters is the angular pixel size and that is determined by the pixel size and focal length, see the left column of the attachment. For a pixel of .00344 mm at 71.2 mm focal length the pixel size is 10" or 291 cycles/mm and 361 pixels/cycle. We are only sampling a real small section of the sensor so overall dimensions are irrelavent unless you want to PH as a unit.

The 10"/pixel is camera or eye level resolution, you must divide by power to get to optic under test angular resolution.

Also, it appears that instead of 2.85 c/p that should be 2.85 p/c or .35 c/p.

Ron,

Sorry to be so slow to reply. I've been a bit busy and I needed to do a lot of checking and managed to confuse myself several times with the mathematics.

Lets recap. The first of the Monarch experiments in #57. The MTF5 was 14.3 cycles/degree. That is 3600/14.3 or 251 arcseconds. Divide that by the 8x magnification and that is 31.5 arcseconds which is the formula you gave in #50.

Edmunds give a nice simple definition of the Nyquist limit (though there may be others).
http://www.edmundoptics.co.uk/technical-resources-center/imaging/resolution/

“The highest frequency which can be resolved by a sensor, the Nyquist frequency, is effectively two pixels or one line pair.”

In the #57 experiment the Nyquist limit is around 16 cycles/degree. Am I right in thinking anything to the right of the Nyquist limit can be ignored as artefact? So that would be (3600/16)/8 or 28.1”. So 28.1” is the theoretical limit of resolution for the sensor derived from the system parameters you entered for the system and 31.5 is the actual observed limit for the whole system. Is that right so far?

I pretty much agree with your calculation for the maximum resolution for the camera in #69 though I had to follow the Imatest formula to get there.
http://www.imatest.com/docs/sharpness/

I calculated 182 cycles/degree or about 20” for two pixels and 10” for one. So that gives a theoretical maximum resolution of 20/8 or about 2.5” (Nyquist Limit) but if we take the experimental value of 2.8 pixels per cycle that becomes 3.5”. A nine fold difference from the fitted MTF of 31.5”. 2.5” or 3.5 "sounds like it's cutting it fine for full aperture testing doesn't it?

So we have a dilemma that your experimental results give a resolution value of 31.5” but the system potentially could resolve 3.5”. Did you use a different focal length for the experiment? The Camera has a range from about 8.9 to 71.2mm or an 8 fold range. Using the Imatest formula for cycles/degree (Pi/180) x (cycles/mm) x (focal length mm). The short axis of the sensor is 6.6mm and a 1920 short axis gives a theoretical 145 cycles/mm. The camera has a potential 22.5 cycles/degree at widest angle or 26.7 arcseconds for the system resolution. If we use the correction of the experimental 2.8 pixels/cycle compared to the 2 pixels/cycle Nyquist limit of 1.4x that would be 31.5”, which is the value we first calculated.

I'd suggest that the two results for the Monarch 8x36 were camera limited and the most the most likely explanation is that the images were taken at the shortest focal length. I've found nothing that would suggest a resolution for the Monarch of 7”. Have I got anything wrong?

David

Hi David,

The bino had a full aperture resolution of 7" measured by standard resolution test. That is 56"/LP at the sensor of 28"/line or 5.6 amd 2.8 pixels respectively which fits your figures.

The bino is the limiting factor, not the sensor. The bino runs out of contrast before the sensor, hence the long, flat tail.

This stuff is WAY over my head but I found this link while trying to learn some more. I am not overly interested in the math but only better understanding what these technical and brilliant minds are talking about. This is mostly for cameras I guess but it is starting to help me more nearly grasp the theme of the last two pages of posts.

http://www.normankoren.com/Tutorials/MTF.html

I just started digesting it (a crumb at a time) but recognized it as helpful or at least interesting and thought I would share with the group.
Jon