Hi Andrew,
I think the algorithm just rescales each row the to the same mean level (on the assumption that the variation in mean from row to row is due to uneven illumination from the lamp or possibly variation in slit width) rather than averaging the columns so the pixel-pixel variation should still be there.