KingofHazor said:
How does it work, though? How is it able to see and do what humans cannot?
If somebody on here works directly in image reconstruction, please chime in and correct me. I know a little bit about image reconstruction in 2D, and I am going to use that information to offer an oversimplified explanation of what I have read in the mainstream (not technical) press. I may be significantly wrong on what these guys are actually doing with these scrolls. I am guessing. I have not read their technical papers.
The field of image reconstruction involves predicting the value of missing pixels. There are a lot of different techniques.
Let's call the most basic technique '2 dimensional tic tac toe'. This assumes a flat image.
Let's imagine a color image sample that has nine pixels, with one unknown in the center. It's a tic-tac-toe board. The value of the center space is the unknown, but we know everything else. We can calculate color gradients for a line of pixels across the center pixel in each direction (E.g., the image is shifting top to bottom from blue to green. Left to right, it is shifting darker to lighter.) We can construct a probability as the average of each of the linear gradient functions at the center pixel. If I know the color value of the 8 surrounding pixels, then I can generate a probability of any particular color for the unknown pixel. My probability for the unknown pixel is essentially a linear extrapolation of a color gradients across the sample. If my sample gets larger, my odds get better. If I have the 16 perimeter pixels outside of my eight on the tic-tac-toe board, then my accuracy is much better.
My analysis gets sharper if I understand the rules of the image. For example, if I know that my image sample is one of 26 letters in monochrome, then I have a pretty solid narrowing of the possible values for my unknown pixel. The math gets a lot easier. If the image has an extended rule set, then it is yet easier. For example, if I know that the language is 1st century Latin, then I have 1) a character set 2) a vocabulary of words to match to the probable character for confirmation and 3) grammatical rules from known samples to use as a further corroboration.
In Herculaneum, you have an added problem.
You're not working in 2 dimensions. You have to take a 3-dimensional image of ink in rolled layers, and mathematically 'roll it out' to generate a flat image. The 3-dimensional imaging problem is solved with medical imaging equipment. You can get a 3-d map of the ink inside the scroll.
Unrolling the scroll is a more complex math problem. The techniques are fundamentally analogous. You add a 4th set of rules: correlations from the flat images of known scrolls, which can be unrolled, and their 'rolled' counterparts. That's how you get to 3 dimensions.
Now, to be clear, I have worked with related problems in 2-d images. I have no direct experience of the 3-dimensional problem or the work that is being done in Campania. I've read a little bit about it, because my family is from the area and I have an interest in the technology. So, my oversimplification has probably distorted what is actually being done.
In the end, if you can get really sensitive 3-dimensional X-ray/IR/CT scans of the rolled scrolls, then the 'unrolling' is 'just math.'
It's REALLY HARD math, but it's 'just math.' For about 200 years, there weren't any good solutions. About 40 years ago, good solutions to the imaging problem started to appear. The computer horsepower has existed for about 10 years.
