- Options
- Login
- Desktop Style
Options
Can VM be written in vowelless Latin?
- Can VM be written in vowelless Latin?
-
RE: Can VM be written in vowelless Latin?
farmerjohn > 11-03-2023, 03:22 PM
(24-02-2023, 04:20 PM)nablator Wrote: You are not allowed to view links. Register or Login to view.There is a small mistake on f23v: qoteotor should be qotcoHtor
You are probably right
this was marked as suspicious and rendered in red.
(Yes, post updates are very overlookable)
(11-03-2023, 10:07 AM)ReneZ Wrote: You are not allowed to view links. Register or Login to view.I have been working in this area off and on, and it came as a bad surprise that a straightforward coordinates conversion between the old and new scans was not going to work.
This issue was so serious, that I had create my own tool for finding subimage in image.
First, the geometry of old and new scans may differ significantly. Folios tend to be skewed near borders, so standard tools (like imagemagick) may be not very effective here. You may have perfect match in the center and complete disaster near edges (see for example right side of 104v).
Next are colors. Newer scans may be darker/lighter, so there can be perfect match in sense of coordinates, but no exact match when comparing pixel to pixel.
And finally, folios can be slightly rotated. Even several tenths of degree can notably improve matching index (no matter how you calculate it).
So my own approach for converting coordinates was the following:
1. Map the whole page to page. Thus get approximate coordinates
2. Map each word rectangle independently, trying coordinates from 1. +- some delta. -
RE: Can VM be written in vowelless Latin?
nablator > 11-03-2023, 05:32 PM
I found another mistake on f80r.35: both qokaiin should be qokain (same mistake in the old TT transliteration, fixed in the newer one).
-
RE: Can VM be written in vowelless Latin?
farmerjohn > 11-03-2023, 07:28 PM
(11-03-2023, 05:32 PM)nablator Wrote: You are not allowed to view links. Register or Login to view.I found another mistake on f80r.35: both qokaiin should be qokain (same mistake in the old TT transliteration, fixed in the newer one).
Thanks, nablator.
Unfortunately, it seems there are still a lot of errors despite various tricks (for example, the search was initially created just to have convenient visual representation of Voynich words and their respective transliterations) -
RE: Can VM be written in vowelless Latin?
ReneZ > 12-03-2023, 05:24 AM
(11-03-2023, 03:22 PM)farmerjohn Wrote: You are not allowed to view links. Register or Login to view.You may have perfect match in the center and complete disaster near edges (see for example right side of 104v)
I immediately had to try this page, of course.
Without making any changes to my tool, it failed from the start, because the very first guess for the first word on the page was not fully inside the search window. The tool keeps track of the offset, once it has found alignment, but it never got there. The quick fix is obvious: increasing the search window. This slowed things down a bit further, but at least I could observe that, indeed, it was not working very well near the right margin.
Only in one case did it miss the word completely, and in numerous cases the box was far too small.
I have some tunings of the tool in mind, to make it keep track in a more intelligent way.
The wrong box sizes will not be easy to fix, but I want to find at least part of the word inside the box, fully automatically.
(11-03-2023, 03:22 PM)farmerjohn Wrote: You are not allowed to view links. Register or Login to view.Next are colors. Newer scans may be darker/lighter, so there can be perfect match in sense of coordinates, but no exact match when comparing pixel to pixel.
This is effectively solved by comparing lightness values and computing the correlation coefficient. This is relatively insensitive to such colour effects.