In case of flipped images, the matching list has to be "translated back":
That is, for each matched end node in skeleton1 (the smaller skeleton),
we translate back the coordinates to the unflipped image and look for the
index of the end nodes with the computed coordinates in the original end
node list.