Earlier workers have noticed that high-pass filtering produces a sharp dark line in precisely the location of the geometrical image of an edge. They proposed using this fact as an aid in measuring linewidth in microscopy but found that the other edge of the line caused significant error. In this paper, I examine that error as a function of normalized linewidth and normalized spatial-filter width and find that it may be limited to ±5% or so, provided that the spatial filter subtends between 0.25 and 0.3× the numerical aperture of the objective and that the linewidth exceeds about twice the resolution limit.
M. Young, "Linewidth measurement by high-pass filtering: a new look," Appl. Opt. 22, 2022-2025 (1983)