Stitch/merge musings and questions

[DavidMillier]

Is the colour moire that visible on an A2 print? Not very nice if it is. I would want to do something about that.

I only have an A3 printer at the moment but I'm thinking about moving up to a P900 and printing 15" square once I have somewhere for such a large printer. The plan is to convert my 8"x6" garden tool shed into an editing studio next year.

 

[DavidMillier]

When you shoot multiple measurements with the same lens and they differ, how do you decide what value to take as representative? The worst, the best, the middle, some average?

 

[DavidMillier]

A friend with a 3d printer has made me a Budgie. I'll be very interested to try it out when I get it.

 

[JimKasson]

When you shoot multiple measurements with the same lens and they differ, how do you decide what value to take as representative? The worst, the best, the middle, some average?

If you’ve done your setup properly, the best is from the image that is best focused, and represents what the lens can do. The average represents what the focusing methodology can be expected to do most of the time.

 

[Erik Kaffehr]

Is the colour moire that visible on an A2 print? Not very nice if it is. I would want to do something about that.

Probably...

Lightroom and other developers can hide color moiré by blurring the color information.

https://www.dpreview.com/forums/post/64621708

This kind of moiré may be rare in real world images, as real word images mostly lack repetitive high frequency detail.

But, check this posting: https://www.dpreview.com/forums/post/64708088

I only have an A3 printer at the moment but I'm thinking about moving up to a P900 and printing 15" square once I have somewhere for such a large printer. The plan is to convert my 8"x6" garden tool shed into an editing studio next year.

 

[ProfHankD]

A friend with a 3d printer has made me a Budgie. I'll be very interested to try it out when I get it.

Enjoy!

Two quick suggestions:

1. Better to use 3 of the 1/2" 1/4-20 metal screws, althought you'll probably have to unscrew/screw one of them to get around the finder bump when mounting the adapter. The 3D-printed screws give a bit much play (I had deliberately made them slightly undersize to allow for printer tolerances).

2. Check that the screws haven't become lose as you shift things back and forth and tighten them as needed.

The design is really a bit strained to be able to take M mount. I'll probably make other versions (e.g., taking Canon EF directly) that can be stiffer, but I went for the most universal mount possible first to make sure there was enough interest to justify the effort....

 

[DavidMillier]

When you shoot multiple measurements with the same lens and they differ, how do you decide what value to take as representative? The worst, the best, the middle, some average?

If you’ve done your setup properly, the best is from the image that is best focused, and represents what the lens can do. The average represents what the focusing methodology can be expected to do most of the time.

Good point. Thank you.

 

[DavidMillier]

Is the colour moire that visible on an A2 print? Not very nice if it is. I would want to do something about that.

Probably...

Lightroom and other developers can hide color moiré by blurring the color information.

https://www.dpreview.com/forums/post/64621708

My Kodak 14n was notorious for colour moire (among many other issues).

My approach was to convert to lab, select the a&b channels and run it through photoshop's dust&scratches filter. Sounds like that is pretty much what LR does without all the steps.

This kind of moiré may be rare in real world images, as real word images mostly lack repetitive high frequency detail.

But, check this posting: https://www.dpreview.com/forums/post/64708088

Colour moire is more obvious than luminance aliasing but the Foveon cameras are good at luminance aliasing. Most fans on the Foveon forum seem blind to it but I feel that once seen it's hard to un-see.

I only have an A3 printer at the moment but I'm thinking about moving up to a P900 and printing 15" square once I have somewhere for such a large printer. The plan is to convert my 8"x6" garden tool shed into an editing studio next year.

 

[DavidMillier]

My first experiments with measuring MTF were done following Erik's suggested method here: https://www.dpreview.com/forums/post/65122260 This involved shooting slant edge test targets, opening in LR, cropping to the desired region of interest and exporting the region of interest as a tiff which was then loaded into MFT Mapper using the "single edge" option and analysed

Today, while playing with the program, I discovered that it can directly open and process raw files (dependencies with dcraw appear to be built in, so nothing special needs to be done other than selecting a raw file and opening it).

I also discovered the "manually select the slant edge" to measure option.

I've been playing with these using the exact same files I shot my first run of tests on. Because the test target is quite small in my frames, I have to zoom in and mark the slant edge region of interest.

Here's the interesting bit: using the raw file directly and doing the ROI selection manually rather than first cropping the file to tiff and letting the software select the ROI appears to yield higher MTF scores than the first method. With the exact same test files as the source.

Example: With my 90mm Sonnar, using the second method yielded a score of 59 lp/mm while the first method gave 23 lp/mm

That seems a bit odd to me.

Any ideas about what is going on?

p.s.

The second method also gives results that are very close to the 40 lp/mm MTF % on the old photodo site for the Sonnar. That may be sheer co-incidence of course as I don't know what their methods are.

--
DPReview gallery: https://www.dpreview.com/galleries/0286305481
Website: http://www.whisperingcat.co.uk/ (2018 - website revived!)
Flickr: http://www.flickr.com/photos/davidmillier/ (very old!)

 

[fvdbergh2501]

My first experiments with measuring MTF were done following Erik's suggested method here: https://www.dpreview.com/forums/post/65122260 This involved shooting slant edge test targets, opening in LR, cropping to the desired region of interest and exporting the region of interest as a tiff which was then loaded into MFT Mapper using the "single edge" option and analysed

Today, while playing with the program, I discovered that it can directly open and process raw files (dependencies with dcraw appear to be built in, so nothing special needs to be done other than selecting a raw file and opening it).

I also discovered the "manually select the slant edge" to measure option.

I've been playing with these using the exact same files I shot my first run of tests on. Because the test target is quite small in my frames, I have to zoom in and mark the slant edge region of interest.

Here's the interesting bit: using the raw file directly and doing the ROI selection manually rather than first cropping the file to tiff and letting the software select the ROI appears to yield higher MTF scores than the first method. With the exact same test files as the source.

Example: With my 90mm Sonnar, using the second method yielded a score of 59 lp/mm while the first method gave 23 lp/mm

That seems a bit odd to me.

Any ideas about what is going on?

Apologies for resurrecting an old thread, but I have a theory:

1. Your lens has non-negligible chromatic aberration. For example, longitudinal CA could cause the red or blue channels to be much less sharp than, say, the green channel.
2. Your LR tiff file was demosaiced by LR to produce an RGB file.
3. MTF Mapper will mix down RGB images to a single-channel luminance image, usually using the channel weights in the ICC profile embedded in the tiff file. Mixing a very blurry red channel into a sharp green channel will inevitably lower the measured MTF50. (A similar phenomenon can occur with lateral CA, where you could effectively have the red edge displaced from the green edge)
4. Confusingly, selecting the "green" Bayer subset in MTF Mapper (preferences dialog) while opening an RGB input image does not discard the red and blue channels. Instead, RGB is mixed down to a luminance image, and MTF Mapper analyses only half of the pixels (that would have fallen on green photosites). So point 3 above still applies, and you end up with low MTF50 values because of CA.
5. Alternatively, if you select the "green" Bayer subset in MTF Mapper, and you open up a raw image file directly (via dcraw / libraw), then the red and blue channels will have no impact on measurements (no reduction because of CA). This very often produces noticeably higher MTF50 values.

Obviously, my theory involves quite a few assumptions, but these assumptions (difference between using step 5 vs 2 through 4) are the ones that will maximize discrepancies.

As a sanity check, you could use MTF Mapper's manual edge selection on the whole tiff file; that should produce the same result you got with the cropped "single edge" measurement. I'd be interested to know if this is not the case.

Thanks for trying out the new MTF Mapper manual edge selection feature, though!

-Frans

 

[Erik Kaffehr]

My first experiments with measuring MTF were done following Erik's suggested method here: https://www.dpreview.com/forums/post/65122260 This involved shooting slant edge test targets, opening in LR, cropping to the desired region of interest and exporting the region of interest as a tiff which was then loaded into MFT Mapper using the "single edge" option and analysed

Today, while playing with the program, I discovered that it can directly open and process raw files (dependencies with dcraw appear to be built in, so nothing special needs to be done other than selecting a raw file and opening it).

I also discovered the "manually select the slant edge" to measure option.

I've been playing with these using the exact same files I shot my first run of tests on. Because the test target is quite small in my frames, I have to zoom in and mark the slant edge region of interest.

Here's the interesting bit: using the raw file directly and doing the ROI selection manually rather than first cropping the file to tiff and letting the software select the ROI appears to yield higher MTF scores than the first method. With the exact same test files as the source.

Example: With my 90mm Sonnar, using the second method yielded a score of 59 lp/mm while the first method gave 23 lp/mm

That seems a bit odd to me.

Any ideas about what is going on?

Apologies for resurrecting an old thread, but I have a theory:

1. Your lens has non-negligible chromatic aberration. For example, longitudinal CA could cause the red or blue channels to be much less sharp than, say, the green channel.
2. Your LR tiff file was demosaiced by LR to produce an RGB file.
3. MTF Mapper will mix down RGB images to a single-channel luminance image, usually using the channel weights in the ICC profile embedded in the tiff file. Mixing a very blurry red channel into a sharp green channel will inevitably lower the measured MTF50. (A similar phenomenon can occur with lateral CA, where you could effectively have the red edge displaced from the green edge)
4. Confusingly, selecting the "green" Bayer subset in MTF Mapper (preferences dialog) while opening an RGB input image does not discard the red and blue channels. Instead, RGB is mixed down to a luminance image, and MTF Mapper analyses only half of the pixels (that would have fallen on green photosites). So point 3 above still applies, and you end up with low MTF50 values because of CA.
5. Alternatively, if you select the "green" Bayer subset in MTF Mapper, and you open up a raw image file directly (via dcraw / libraw), then the red and blue channels will have no impact on measurements (no reduction because of CA). This very often produces noticeably higher MTF50 values.

Obviously, my theory involves quite a few assumptions, but these assumptions (difference between using step 5 vs 2 through 4) are the ones that will maximize discrepancies.

As a sanity check, you could use MTF Mapper's manual edge selection on the whole tiff file; that should produce the same result you got with the cropped "single edge" measurement. I'd be interested to know if this is not the case.

Thanks for trying out the new MTF Mapper manual edge selection feature, though!

-Frans

Thanks for your thoughtful comments!

Erik

 

[DavidMillier]

Hi Frans

Thanks for the ideas, very helpful.

I'm a complete amateur at testing - having come back to this as a result of your post, I find my notes are a mess and I can't even find the spreadsheet I put together to record all the results. I've still got some random files in folders but I don't know what they are.

I need to start again.

I switched to one of your supplied charts (with the little slant edges blocks all across the frame) at some point but I seem to have mislaid those results as well (I suspect in a fit of pique, I may have deleted them!).

Thinking back, I think the primary issue with starting using the tool is the absence of any kind of reference result to replicate as a check that everything has been done correctly. There is an opportunity for to sell a test reference lens with a known performance shipped with a "factory" MTF score for home calibration purposes

Because I have been testing my legacy vintage lenses, it's been difficult to know what kind of results to expect. My latest plan is to acquire a single modern native mount lens with a good reputation to act as my own personal reference and allow me to compare my vintage lenses to. If that lens shows similarly oddball results to some of the results I've got so far, at least I could be confident that problem lies with my test technique rather than the lenses.