From my experience, the fuzzy matching works well enough to handle picking out text or icons on interfaces with semi-transparent backgrounds that change on scrolling (e.g., due to a parallax effect). It's pixel-based and not scale-invariant, though, so it won't automatically handle multiple zoom levels/resolutions.