The solid lines all refer to the same final pixels, the 4:1 downsized ones, so their results are as measured. The dashed original line is there for reference.
Yes, it's the dashed original, in particular, that I don't understand (the MTF curve seems way too high).
Did you resize the image 4:1 using the various methods?
MTF Mapper never gets confused, if anything it's operator error But in this case it looks like you are using the original edge at its native resolution so that's where the discrepancy comes from. And you are probably unknowingly adding a little sharpening somewhere in your workflow, because the MTF50 value in cy/px looks high. Have you tried running Imatest on the cropped tiff I provide there?
I was using the original image as supplied in your link ... which corresponds to the dashed line, correct? I didn't add any sharpening - if I had my MTF result might have been a bit nearer yours, but as it is the Imatest MTF result was much lower, as you can see.
Here is the MTF with bicubic downsized by 4x (which actually leaves the image not really big enough to run Imatest on). This has added quite a bit of sharpness - or possibly the Imatest result is not reliable because the sample has become too small. At any rate the result is not too far off your bicubic result.
So it's really your dashed curve that I don't get.
Cheers,
Robert