Extremely touchy software when attempting to align the test image
Status: Beta
Brought to you by:
fvdbergh
Menu
▾
▴
Won't produce a good lensprofile except by accident. Move the chart a tiny amount and the orientation numbers jump around like crazy. My last attempt gave me a big red circle with a line through it. "Fiducials not found, probably incorrect chart type." But everything is right there in the image. Well...what chart was I using? YOUR CHART. I guess that explains it.
I have the chart inside a frame with a glass covering to hold the chart flat. Maybe it's any reflections that are driving the software nuts. But the software is producing garbage.
Well that sounds frustrating!
I'd be happy to help you obtain better results. Some preliminary observations:
1. I would not recommend covering the chart with any material that is likely to produce specular reflections, such as glass or plastic lamination. Such reflective surfaces are prone to producing nonuniform illumination, which are inevitably going to make it harder to obtain good results.
2. Keep in mind that lens profiles produced by MTF Mapper are rarely going to look like those published by the manufacturers. They typically publish simulated results; real lenses do not usually have the radial symmetry that would guarantee a good "lens profile" plot. For example, a lens with some tilt might have good MTF50 values going from the centre to the 2 o'clock direction, but much worse values going from the centre to the 8 o'clock direction. This direction-dependent variability shows up as a large standard deviation band (the shaded section around the plotted line), but it also means that the "average" plotted line is unlikely to be correct. See Fig 3 of this paper. Fig 7 of the same paper illustrates just what real world lens copy-to-copy variation looks like.
3. You don't have to rely on just the chart orientation estimation to obtain good chart-to-sensor-plane parallelism. I know that the chart orientation estimation feature is not quite perfect, especially for lenses that have very different parameters compare to the lenses I had access to while developing the software (i.e., I could only validate against the lenses I have). What I do recommend is that you use the mirror method, i.e., hang a small mirror (just the bare glass part) over the chart, and adjust the chart+mirror orientation until you see the lens looking right back at itself in the mirror. That should give you good enough alignment, but please let me know if this is still inadequate.
4. Chart orientation requires a good estimate of the lens focal length. If you run MTF Mapper through the GUI, and your input file contains EXIF tags (TIFF and JPEG, but also in the raw formats supported by MTF Mapper through LibRaw / dcraw), then that value should be extracted automatically. Running the command line version of MTF Mapper requires you to provide this value.
Regardless, I'd be happy to help debug this, if you are willing. Please send me a sample image or two (fvdbergh@gmail.com) if you'd like me to offer better suggestions.
Regards,
Frans