- Options
- Login
- Desktop Style
Options
Color Deconvolution / "Depainting"
- Color Deconvolution / "Depainting"
-
RE: Color Deconvolution / "Depainting"
oshfdk > 10-12-2024, 05:47 PM
(10-12-2024, 05:33 PM)Mauro Wrote: You are not allowed to view links. Register or Login to view.A great job, wow. I can't wait to see the whole processed VMS (and my eyesight will greatly benefit from it!).
Thank you! I've actually embarked on a huge self-educational project, where I'd like to train a full point-based OCR system for Voynichese. I'd like it not only to be able to identify Voynichese characters, but also provide specific information about endpoint coordinates for various penstrokes, like EVA r/s/n flourishes. Needless to say, I have no idea how modern OCR systems work and reading a few texts on convolution neural networks left me head-scratching, but I think if I proceed slowly step by step I might get somewhere.
The images that I've shown here are the side product of the first step in OCR: removing unnecessary elements and improving the contrast of the text. -
RE: Color Deconvolution / "Depainting"
BessAgritianin > 11-12-2024, 09:07 AM
oshfdk,
For long time I have tried to read the explanatory marginalia, which have been hidden, either by paint, or by deletion, or by writing with transparent ink (?).
One of the places, I lost most time is the top of Q1f1v. There is an obvious writing, but I have not succeeded in reading it completely.
You have not decoloured it why?
Could you try it with your methods?
In the attachment I hope to succeed this time and demonstrate one of my results, using relief distortion.
br: Vessy -
RE: Color Deconvolution / "Depainting"
oshfdk > 11-12-2024, 09:44 AM
(11-12-2024, 09:07 AM)BessAgritianin Wrote: You are not allowed to view links. Register or Login to view.You have not decoloured it why?
I'm not manually picking what to decolour, I only provide the model with a few samples of color and ink areas and the model tries to learn their properties and then apply them uniformly on the whole MS. Since I trained the model just on part of a few images mainly from the first part of MS, is doesn't work well on combinations of paints that it had no data on (e.g., it turned out to be not very good when depainting the water in balneological sections).
Looking specifically at the flower top, I'm not even sure it wasn't just painted (hatched) over with the same ink as the text. The hue is a tiny bit off, if you look at the hue channel in GIMP, but this looks more like text ink than anything else to me. If this is text ink, there is nothing I can do to depaint it using the models that process local pixel values.
Quote:Could you try it with your methods?
Most probably it won't work with visible light TIFFs, even if there is indeed some writing underneath. However, we are talking about f1v, which is the verso of f1r, for which we do have access to some MSI images. Unfortunately there are no TX (shine through) images for You are not allowed to view links. Register or Login to view. (as opposed to almost every other folio in the MSI set), but maybe something can be extracted from the existing images. I'll have a look. -
RE: Color Deconvolution / "Depainting"
ReneZ > 11-12-2024, 11:54 AM
(11-12-2024, 09:07 AM)BessAgritianin Wrote: You are not allowed to view links. Register or Login to view.One of the places, I lost most time is the top of Q1f1v. There is an obvious writing,
People have argued in the past that there seems to be something there, but it is far from obvious.
For what it is worth, I have looked, but I have never seen anything convincing there. -
RE: Color Deconvolution / "Depainting"
BessAgritianin > 11-12-2024, 01:58 PM
(11-12-2024, 11:54 AM)ReneZ Wrote: You are not allowed to view links. Register or Login to view.People have argued in the past that there seems to be something there, but it is far from obvious.
For what it is worth, I have looked, but I have never seen anything convincing there.
Maybe it is not obvious to You, but also if someone has not seen anything convincing- that does not mean there is nothing.
I even can read some part of an explanation there.
So my question to oshfdk is may it be tried to be discoloured also?
br: Vessy -
RE: Color Deconvolution / "Depainting"
oshfdk > 11-12-2024, 03:01 PM
I never noticed that the twining sprouts seem to go right through the stalk/trunk on f35v. I wonder what this could signify. (Sorry for the off-topic, I'm just going through images to check that they are OK.)
-
RE: Color Deconvolution / "Depainting"
oshfdk > 11-12-2024, 03:32 PM
(11-12-2024, 01:58 PM)BessAgritianin Wrote: You are not allowed to view links. Register or Login to view.So my question to oshfdk is may it be tried to be discoloured also?
I can't separate layers of the same colour. I've tried looking at MSIs of the recto face, visually there is almost nothing there, you can have a look yourself. For the following image I cropped the corresponding part of You are not allowed to view links. Register or Login to view. from all MSI images, flipped horizontally and overlaid the flower with the center removed. It was very easy to match it more or less exactly, because the leaves from You are not allowed to view links. Register or Login to view. are clearly visible on f1r.
Until we have an MSI of You are not allowed to view links. Register or Login to view. or at least TX MSI of f1r, I think it's impossible to determine if there is some inscription there or just scribbles. -
RE: Color Deconvolution / "Depainting"
oshfdk > 11-12-2024, 03:38 PM
(11-12-2024, 01:58 PM)BessAgritianin Wrote: You are not allowed to view links. Register or Login to view.Maybe it is not obvious to You, but also if someone has not seen anything convincing- that does not mean there is nothing.
I even can read some part of an explanation there.
You can try doing what @Searcher did in this post: You are not allowed to view links. Register or Login to view.
If there is a specific shape within this flower that you want checked, it's possible to try training a model on it, but I'll need a specific shape to separate points that are clearly part of the feature and points that are part of the background. -
RE: Color Deconvolution / "Depainting"
Koen G > 11-12-2024, 05:54 PM
(11-12-2024, 03:01 PM)oshfdk Wrote: You are not allowed to view links. Register or Login to view.I never noticed that the twining sprouts seem to go right through the stalk/trunk on f35v. I wonder what this could signify. (Sorry for the off-topic, I'm just going through images to check that they are OK.)
This has been discussed in this huge thread: You are not allowed to view links. Register or Login to view. (one of the last posts happens to be one where I post a random late parallel).
Apart from speculation, I don't remember any conclusion being reached.