(17-12-2024, 07:24 PM)Koen G Wrote: You are not allowed to view links. Register or Login to view.Do you know what kind of scan would be required to get the best contrast between ink and paint? Or to even reveal ink under the paint?
This is a very interesting question. I suppose, MSIs should be more than enough for this task, but doing MSIs of the whole codex is a complex and potentially harmful procedure, as far as I understand.
Given that we have MSIs for a number of pages with ink and paint, it should be possible to take wavelength metadata from MSIs and compute which narrow band optical filter (of wavelengths that would work with a normal camera, not IR or UV) would produce the maximum separation between ink and everything else. Most likely it won't be useful for visual inspection, since narrowband filters are darkish, but modern camera sensors are very sensitive, so it should be possible to snap the filter onto a decent consumer-level camera and take a few pictures.
Note that I'm an amateur in this. I have used narrow band filters for non-professional photography, they do produce very cool effects, but for this task maybe there is a better professional solution. E.g., if there is a way to combine a few narrow band filters (via double exposure?) the result could be even better.
On the other hand, what is the value of identifying a new label or two under paint, if we have access to the whole corpus of text and no luck with deciphering it so far.
Labels probably not so much. Plaintext on the other hand...
(18-12-2024, 05:55 AM)Koen G Wrote: You are not allowed to view links. Register or Login to view.Labels probably not so much. Plaintext on the other hand...
"for the key see the back of the cover"
I've tried to train a separate instance using visible parts the apparent text on the leaf, that's what it produced:
[
attachment=9577]
There is not much difference with the generic model, but from the look of it I wouldn't exclude the possibility of "eod an on" or even "sod an on". I'm not suggesting these readings are more likely, I'm just throwing them in the mix.
Just for clarity, what was it trained on. Left top is the original TIFF, right top is the original TIFF with training sample overlay, the white points were used as the background sample, the pink points as the ink sample. Below is the result of applying the trained model to the whole area.
[
attachment=9578]
For what it's worth, with Gabriel Landini's filter (green color separated then subtracted) and the local contrast filter in ImageJ I got:
[
attachment=9579]
This is one of the cases where the paint is not hiding the writing completely. Maybe only X-rays would make writing visible through a thick opaque paint.
I think this is a red herring, but some of the models (linear svc and rbf svc) consistently detect some vague shape in the right part of the root on f2r, which could be a couple of letters. More likely just a peculiar way several paintbrush strokes connected together.
[
attachment=9580]
I won't trust my eyes for this kind of analysis, but I think I can see a small EVA-l like shape in approximately the same area. I've tried to boost it with some color correction.
[
attachment=9581]
Speaking of the picture that I uploaded, I was probably overthinking.
I have seen something like that:
You are not allowed to view links.
Register or
Login to view.
Where the first word is TIC and the second who knows.
But it's just the
pareidolia, I guess
(18-12-2024, 06:32 PM)Rafal Wrote: You are not allowed to view links. Register or Login to view.But it's just the pareidolia, I guess
There is something, plainly visible on You are not allowed to view links.
Register or
Login to view. even in unprocessed images. Maybe "Ila" or
ka.
(18-12-2024, 06:32 PM)Rafal Wrote: You are not allowed to view links. Register or Login to view.Speaking of the picture that I uploaded, I was probably overthinking.
I have seen something like that:
You are not allowed to view links. Register or Login to view.
Where the first word is TIC and the second who knows.
But it's just the pareidolia, I guess
When a shape is deeply embedded in the drawing, I think, this is not a question of detecting something, but a question of intent of whether, say, "ccciv" represents the petals of a flower or number 304. A sequence of pen strokes accompanied by "as shown above the petals are small and curly" is semantically different from absolutely identical sequence of strokes with "there is a number hidden in this image, can you find it?" written underneath
