I've written before a bit about "transparency holograms" and how they might be useful as an archival medium for text. Here I want to do some calculations on that.
Let's suppose we're printing out on both sides of a 4-mil (102μm) Mylar transparency foil, using the simple grille pattern where all but one pixel is blacked out in each N×N square; I'll take N to be 7 in this case, and assume 600dpi, so the grille has a hole spacing of 7"/600 = 296μm, or 85.7 dpi. The other side of the foil has the pixels of 49 different 85.7 dpi pages interleaved. (For the moment I'll assume this is for a viewing distance of infinity.)
If we use a 6-pixel-tall font for text, which works out to 5.04 points at 85.7 dpi, each of the 49 pages has space for 157 rather long lines of text, about 208 characters each if we assume an average of 3.5 pixels. (I don't remember if this is the right average from bible-columns.png.) That's about 1.6 megabytes of text on the transparency, and it should all be visible to the naked eye, hopefully.
Do the viewing angles work out? The angle between two pages needs to not be impracticably small, and the angle for the most extreme left, right, top, or bottom page needs to not be impracticably large. Mylar's index of refraction is about 1.6, so those 101.6 microns are, optically speaking, closer to 63.5 microns. But our 600dpi page has 42.3μm pixel spacing, which means that we have to go 25° to the left, right, up, or down to see the next pixel in that direction: 20 cm to the side at a viewing distance of 50 cm. The refraction linearizes this somewhat, so the next page is closer to being 20 cm further to the side than to being 25° further; about 60cm to the side gets you to the third page. So we can do 7×7 without extremely distorted viewing angles—barely! At the extreme corners of viewing angle, the page is foreshortened by 36% in both X and Y dimensions.
So at 600dpi on heavy-duty transparency film, it's barely doable; at 1200dpi it's eminently doable, with more like 6 megabytes of text per page; at 300dpi you would need thicker transparency film.
14-mil (356μm) and 7-mil (178μm) Mylar sheets are sold as art supplies for making stencils, pre-frosted on one or both sides, even. This would work even with a 300dpi printer, but necessarily with a wider grille spacing. "Matte Dura-Lar" is one brand name for double-frosted Mylar. "Drafting film" is another way this kind of thing is sold. I suspect that clear film will work better, though. Grafix, the manufacturer of Dura-Lar, claims that neither inkjet nor laser will print on Dura-Lar, but other people claim success with inkjet, while claiming that the laser-printer heat curls up Mylar.
A related idea: shrink film, like Shrinky-Dinks, might enhance the resolution of a printer at the same time as thickening the film; it shrinks 2:1 when heated, which should quadruple its thickness, initially 254μm. But I don't know if it's available in transparent, and it's made of polystyrene, which is apparently not archival-safe if exposed to ultraviolet or, maybe, oxygen. However, it seems that polystyrene's better-knn problem is that it's so stable that it can last a long time after being discarded, resisting biodegradation and photodegradation for up to centuries. (Obviously you can't put shrink film through a laser printer.)
Mechanically, though, polystyrene is far, far more fragile than PET. PET stick-on films are used to make windows smash-resistant and even mildly bullet-resistant.
At the other end of the spectrum, polycarbonate ("bulletproof glass") can be recovered from CDs, and is, ounce for ounce, less fragile than almost anything, although it has a weakness for ammonia and other alkalis. Melting CDs also causes them to shrink, just like polystyrene shrink film; but they're optically clear. It's also an archival-stable plastic.
I snarfed rosettaproject_zul_gen-1.pdf (SHA1 296b27e22a720a8d3c10fa971d262f798b27906d) from https://archive.org/details/rosettaproject_zul_gen-1 and converted it first to PS with pdf2ps, then to 100dpi PNGs. GhostScript by default just does point sampling rather than antialiasing, which is shitty and loses parts of letters entirely when rendering bitmaps at low resolution, so I antialiased by brute force with ImageMagick:
time gs -dNOPAUSE -r300 \
-sOutputFile=rosettaproject_zul_gen-1-large-p%d.png \
-sDEVICE=pnggray -dBATCH rosettaproject_zul_gen-1.ps
time for x in rosettaproject_zul_gen-1-large-p*; do
mogrify -scale 827x1169! -threshold 90% "$x"
done
827×1169 is from units A4paper telling me A4 is 210mm by 297mm, and
that those work out to be 827 and 1169 pixels respectively at 100dpi.
The ! is to indicate that the images are to be stretched and
squeezed as necessary to make them fit, because by default ImageMagick
preserves aspect ratios and doesn't fit the box. The -threshold
option is to convert the image to bilevel, while dilating the
letterforms somewhat to keep any lines from getting lost.
Anyway, then I use the interleave.py I just wrote using PIL;
unfortunately this is only 5 pages, so to get to 3×3, I had to
duplicate some.
time ./interleave.py \
rosettaproject_zul_gen-1-large-p1.png\
rosettaproject_zul_gen-1-large-p2.png\
rosettaproject_zul_gen-1-large-p3.png\
\
rosettaproject_zul_gen-1-large-p4.png\
rosettaproject_zul_gen-1-large-p5.png\
rosettaproject_zul_gen-1-large-p5.png\
\
rosettaproject_zul_gen-1-large-p5.png\
rosettaproject_zul_gen-1-large-p5.png\
rosettaproject_zul_gen-1-large-p5.png
mv interleave-output.png bible-zulu-interleaved.png
interleave.py took like 3½ minutes to do this, and the result looks
okay.
(Now I guess I need to turn that into a PDF so I can print it, plus printing the grille.)
My theory on the numbers on this prototype is that, even if the number of pages changes, the size of the composite pixels shouldn't; there should still be about 86 of them per inch in each direction. This image has 100 of them per inch in each direction, and only 9 pages; this should be less challenging in the dimensions of legibility, viewing angle delta, brightness, and maximum viewing angle.
A thing that surprised me about this, but shouldn't have, is that some of the text is readable without the grille. In particular, since page 5 is repeated 5 times, it's quite readable despite the other four page images interfering.
I downloaded Murray R. Spiegel's "Mathematical Handbook of Formulas and Tables" from https://archive.org/details/MathematicalHandbookOfFormulasAndTables (SHA1 efc83d86ae251e4ef7bf6ec852caeb95708ba98d) and extracted pp. 27, 28, 30, 31, 37, 38, 39, 40, and 43 with Evince's "print; range; pages" to a PDF file. Then I processed this into PNGs:
time gs -dNOPAUSE -r300 -sOutputFile=spiegel-select-p%d.png \
-sDEVICE=pnggray -dBATCH ~/Documents/spiegel-select.pdf
time for x in spiegel-select-p*; do
mogrify -scale 827x1169! -threshold 70% "$x"
done
The 90% threshold came out shitty because this book contains gray halftoning with super important black text in it, but a 50% threshold washes out the text badly. 70% was better than either 66% or 75%, although it still has some problems.
time ./interleave.py spiegel-select-p*
mv interleave-output.png spiegel-interleaved.png
It seems to have worked pretty well.