On Sat, 10/26/2024 7:49 AM, Java Jive wrote:
Some here may remember that over the last few years I've been ploughing through a family history project involving thousands of documents from two branches of our family going back many generations to at least the reign of Queen Anne, and that I've been scanning big documents in sections and stitching those together with photo panorama stitching software
Undoubtedly the best for this has been Image Composite Editor or ICE (Windows). However, this failed me for the last big family tree, a Holroyd (Yorkshire family) tree of 3 x 19 A4 sections, which surprised and disappointed me, as previously it had handled a bigger 5 x 17 sections tree pretty well.
Accordingly I've spent the last week in Hugin (Linux) creating control points between neighbouring sections. However, when I came to try to create the panorama, it spewed out errors that I've not seen before, and about which I've not, so far in a brief search, found out much about online, as follows:
First, a dialogue box is shown twice in quick succession ...
PTBatcherGUI
An assertion failed!
/usr/include/wx-3.0/strvararg.h(462): assert "(argtype & (wxFormatStringSpecifier<T>::value)) == argtype" failed in wxArgNormalizer(): format specifier doesn't match argument type
Backtrace:
1 wxEntry(int&, wchar_t**)
2 __libc_start_main
... and thereafter the batch processing log then shows multiple errors similar to ...
enblend: note: seam-line end point outside of cost-image
enblend: note: contour #1 of 1, segment #1 of 1, vertex #1583 of 1590
... and ...
unable to run Dijkstra optimizer
Note: PCs tried are Dell Precision M6800 32GB RAM, Dell Precision M6700 24GB RAM, and Dell Precision M6300 8GB RAM, all running Ubuntu 22.04.5 LTS. Hugin Panorama Creator, Help, About on the first and best PC gives Hugin version as 2023.0.0.d88dc56ded0e, but also gives an error "File couldn't be loaded: '/usr/share/hugin/xrc/data/COPYING.txt'
All three attempts to stitch exhibited the same above errors.
Any ideas?
Just a word of warning. Paper stretches and shrinks with humidity.
It's fine if you scan it right away after it has come from a
printer. But a year later, there can be a 2% error in scale between
horizontal and vertical, and the scale error isn't even constant.
There can be blotchy scaling errors going on.
ICE cannot really handle this, like if the image has suffered too much
at the hands of the paper fiber. I've had things joined together
that have sections that don't align at all. It can only do so much,
when the objects don't actually overlap and can't actually correlate.
it can make some features align really well, but when the gravy is
too wavey, it just has to give up and pave it. If you did a math transform
to distort the text severely, I bet the glyphs could be joined, but
a human would gag at the overall effect (too wavey, too gravy). To
the user, the distortions necessary to make it join, would have
no analog in the real world. The observer would not be able to tell
why the panorama was this bad.
I bet an AI could fix this. At some cost to your sensibilities :-)
It could quite easily look like crap and the AI would be happy
about the whole thing.
I don't know what the error mean, but the long and the short of it,
is your source is damaged. Start with a 2x2 and force it to complete
a small task. Then examine the result carefully, and see just how bad a
job it did with a 2x2. When you do a 3x19, things have to be
"very healthy" for this to work. This is why scanning from film stock
is about the best source you can think of, with the panorama shot from
a panorama camera stand. Activities such as buying a "map book" at the
store, cutting out the paper pages, scanning and joining, that doesn't work.
It doesn't work if you wait a year, before doing the scanning step.
Paul
--- MBSE BBS v1.1.0 (Linux-x86_64)
* Origin: A noiseless patient Spider (2:250/1@fidonet)