[split] Jorge's Retracer Thread: darker ink, retracing of text and drawings

[Jorge_Stolfi]

By the way, great caution needs to be used when overlaying the entire width of the page. The parchment seems to have been in a slightly different orientation between different scans

Indeed the mapping from the Beinecke image of You are not allowed to view links. Register or Login to view. to match the UV image of You are not allowed to view links. Register or Login to view. required a slight rotation and stretching.  The projective transformation that I found by matching four key points (hole corners) was

    [  1.0370  4560.6296  -159.9366 ]
    [ -0.0000    -1.9335    -0.0387 ]
    [ -0.0000     0.0268     1.8474 ]

which means 

  Xout = ( 4560.6296 - 1.9335 * Xin + 0.0268 * Yin ) / 1.0370
  Yout = ( -159.9366 - 0.0387 * Xin + 1.8474 * Yin ) / 1.0370

(Unless I am switching the X and Y here, maybe.) This is actually an affine transformation (linear map plus a translation).  The "in" coordinates are measured from the upper right corner of the Beinecke image, while the "out" coordinates are measured from the upper left corner of the UV image after

Code:

-crop '6540x4300+1067+1890' -rotate -90

(how flat or bumpy (or stretched?) it was in certain areas). As far as I can tell, there is no way to overlay the entire page in a way that lines up with everything perfectly.

Indeed the Beinecke images were apparently taken without any attempt to keep the vellum flat.  This can be seen on the Zodiac pages where the larger circles are slightly squeezed sideways while the smaller ones are almost perfectly round.  

Whereas the folios were at least partly clamped to some flat surface when the Multispectral images were taken.

Thus the mapping above is expected to get visibly wrong near the binding fold.

Still, the holes -- especially the two at the bottom right -- got matched with at most one penstroke-width of error.  That is my estimate for the maximum error of the mapping above.

It is quite possible that a more precise mapping, that accounted for the curvature of the Beinecke image, would get those two ls to coincide more closely with the big "question mark".  I don't dismiss the possibility that they were what got the Repairman to draw that sign.  However I still maintain that there is real ink there, not just the seethrough of the ls.

It's better to align a smaller area when looking for a specific mark as it's less likely that these factors will cause deviations.

But doing it this way increases the risk of aligning things that are not really aligned.  Are those two ls really aligned with the question mark, as in your previous alignment image, or are they in fact 0.5 mm to the left?

For example, here are our overlays compared to the multispectral scan for the end of "gasmich". Note how a tiny alignment difference actually makes quite a large difference.

The difference between your alignment and mine, in this area, seems to be about the width of one broad pensroke in the horizontal direction, and a small fraction of that in the vertical direction.  Which matches my estimate of the error, above.

These discrepancies would make a big difference when evaluating the influence of seethrough on a specific glyph.  But when one considers a whole word or line, these discrepancies do not seem relevant: overall, the "anchiton" text does not seem to be aligned or influenced by the seethrough.

Moreover, as I mentioned, the seethrough would be clearly indentifiable as such when looking at the real vellum, because it would be a bunch of fuzzy shapes, clearly on the other side of the vellum.  I can't believe that the Repairman would have mistaken them for remains of the original text, no matter how faint. Except maybe in one or two cases, like the "question mark".

All the best, --stolfi