f116v squiggle from the multispectral images

[oshfdk]

This is not an erasure, as far as I can see from the raw TIFFs, it's just that the model seems picking on contours around ink. So, two black lines are not the frown and the nose per se, but the area immediately below the nose and upwards from the frown. If you put the nose where the white nose shape is and the frown immediately below the black frown line, they are roughly at the same positions as with the visible light scan.

Good question is how exactly the model picks the areas surrounding ink. It won't pick visible ink, because there is no visible ink at the bottom right of You are not allowed to view links. Register or Login to view. where is was trained. This model processes the image in 2x2 cells, which is too small to explain these artifacts.

It's possible that at some wavelengths there is substantial glow (UV causes some luminescence) which is leaking some light into the vellum around ink spots. I've been reading and watching a bit of introductory material on MSIs, if I understood it correctly one of the reasons some hidden ink can be visible in UV because UV can reach particles of ink buried inside medium and make them glow, which lights up some small area around them.

If this is the case, the bottom part of the glow of the frown could be hidden under red paint and didn't register, that's why there is only one black line.