15-12-2025, 07:15 PM
More hallucinations in Voynichcolor™® from page f17r, using the Lazarus 2016 multi-spectral images! Featuring another possible example of "imperfect retracing" (B).
This clip spans the left end of lines 1-3 of page f17v. \as usual, the text shows clues of original traces (Rt0), the first genenral restoration pass (Rt1) and at least one subsequent retouching pass (Rt2).
The image is a composite derived from three multi-spectral images taken by the Lazarus team in 2016, with wavelengths 365 (UV), 625 (red), and 870 (IR). The processing included scaling the samples in each image from integer range {0..maxval} to the float range [0_1]}, where {maxval} is the sample value seen on the Spectralon sample included in the image. Then for each pixel, the 12 samples from the 12 narrow-band images obtained with both light sources (files "MB*" and seq numbers 001 to 012) were scaled so that their average value was 0.4. (This step largely eliminates variations from pixel to pixel due to vellum warping and fine-scale relief.) Then the images with the three selected wavelengths above were combined as the Y, U, and V channels of an image in YUV color space, which was then converted to RGB. At various steps the images were normalized so as to achieve maximum contrast in the area covered by writing anda bit of background around it.
In this image, original traces are dark but fading blue, Rt1 traces are purple, and Rt2 traces are red with purple spots. The colors are modified somewhat in areas (X) where the vellum color is affected by bleed-through of the green paint on the other side, f17v. In those areas, all trace colors on this image are shifted towards red, so that Rt0 looks ike Rt1 elsewhere, Rt1 looks like Rt2, and Rt2 is bright red.
(A) The loop of the f is mostly original (Rt0). (B) The plume is Rt0 except for a semicircular imperfect retracing by Rt1. (C,D,E,F,G,H,I,J) The body of glyph and the beginning of the tail are Rt1 or Rt2, but the distal half of the tail is still Rt0. (K) The body of this glyph is Rt1 or Rt2, but the whole tail is still Rt0 and faint, so it could be transcribed as a. (L) Most of the plume is Rt2, but the tip is Rt0.
(N) This glyph, originally o, was slightly deformed by Rt1 so that it could be confused with a y with faded tail.
(P,Q) The C of the Sh is Rt2, while the h is Rt1. (R ) The C of Ch is Rt2, the h is Rt1. (S) The body is Rt2, and the plume was wholly retraced by Rt1 or Rt2. (T) The body and half of the plume of this r is Rt2, the rest of the plume is Rt1, traced the wrong way (clockwise). (U,V) Bit of Rt2 on Rt1 glyphs.
Here is the same area from the Beinecke Library 2014 scans:
Frankly I don't see much gain in using the multi-spectral images here. Al the details I can see in the pseudocolor image I can see also on the BL image. The resolution is a bit better, and maybe it will be easier to automate the analysis with the Bayes method I described before. We will see...
All the best, --stolfi
This clip spans the left end of lines 1-3 of page f17v. \as usual, the text shows clues of original traces (Rt0), the first genenral restoration pass (Rt1) and at least one subsequent retouching pass (Rt2).
The image is a composite derived from three multi-spectral images taken by the Lazarus team in 2016, with wavelengths 365 (UV), 625 (red), and 870 (IR). The processing included scaling the samples in each image from integer range {0..maxval} to the float range [0_1]}, where {maxval} is the sample value seen on the Spectralon sample included in the image. Then for each pixel, the 12 samples from the 12 narrow-band images obtained with both light sources (files "MB*" and seq numbers 001 to 012) were scaled so that their average value was 0.4. (This step largely eliminates variations from pixel to pixel due to vellum warping and fine-scale relief.) Then the images with the three selected wavelengths above were combined as the Y, U, and V channels of an image in YUV color space, which was then converted to RGB. At various steps the images were normalized so as to achieve maximum contrast in the area covered by writing anda bit of background around it.
In this image, original traces are dark but fading blue, Rt1 traces are purple, and Rt2 traces are red with purple spots. The colors are modified somewhat in areas (X) where the vellum color is affected by bleed-through of the green paint on the other side, f17v. In those areas, all trace colors on this image are shifted towards red, so that Rt0 looks ike Rt1 elsewhere, Rt1 looks like Rt2, and Rt2 is bright red.
(A) The loop of the f is mostly original (Rt0). (B) The plume is Rt0 except for a semicircular imperfect retracing by Rt1. (C,D,E,F,G,H,I,J) The body of glyph and the beginning of the tail are Rt1 or Rt2, but the distal half of the tail is still Rt0. (K) The body of this glyph is Rt1 or Rt2, but the whole tail is still Rt0 and faint, so it could be transcribed as a. (L) Most of the plume is Rt2, but the tip is Rt0.
(N) This glyph, originally o, was slightly deformed by Rt1 so that it could be confused with a y with faded tail.
(P,Q) The C of the Sh is Rt2, while the h is Rt1. (R ) The C of Ch is Rt2, the h is Rt1. (S) The body is Rt2, and the plume was wholly retraced by Rt1 or Rt2. (T) The body and half of the plume of this r is Rt2, the rest of the plume is Rt1, traced the wrong way (clockwise). (U,V) Bit of Rt2 on Rt1 glyphs.
Here is the same area from the Beinecke Library 2014 scans:
Frankly I don't see much gain in using the multi-spectral images here. Al the details I can see in the pseudocolor image I can see also on the BL image. The resolution is a bit better, and maybe it will be easier to automate the analysis with the Bayes method I described before. We will see...
All the best, --stolfi