I assume many of you host a DMS such as Paperless and use it to organise the dead trees you still receive in the snail mail for some reason in the year of the lord 2023.
How do you encode your scans? JPEG is pretty meh for text even at better quantisation levels (“dirty” artefacts everywhere) and PNGs are quite large. More modern formats don’t go into a PDF, which means multiple pages aren’t possible (at least not in Paperless).
Discussion on GH: https://github.com/paperless-ngx/paperless-ngx/discussions/3756
Well, what kind of lossy compression? JPEG?
IME, JPEG looks quite terrible for text documents -even at q=95.
Yeah just jpeg. Always comes out perfectly legible.
@Atemu
I just use grayscale PNGs, myself. optipng usually takes them down to a decent size.
@Saigonauticon
Hmm, I’m using grayscale PNGs as my baseline here. A 150dpi scan is about 1.3MiB.
A (for the purpose of text documents) similar quality WEBP is about 1/4 of that.
You could also try adjusting the contrast a bit. I use an app called Genius Scan, which increases the contrast of the scanned image to reduce the number of bits needed per pixel. This reduces the size of the file quite a bit, although it obviously isn’t a true representation of the scanned document. The TextCleaner imagemagick plugin looks like it’s doing something similar.
@Atemu
Webp is much better, as long as your target reader(s) support it.
Yes, as I said.
As also mentioned in the post, I need a solution for multiple pages and an image (no matter what format) only represents a single page and WEBPs don’t go into PDFs.
@Atemu
There’s not really a magic bullet here. The current answer is to prepare a PDF outside of paperless and feed it in: https://github.com/paperless-ngx/paperless-ngx/discussions/367
mpflanzer on that Issue is working on a file merging feature, but it’s not ready yet.
That’s nice and all but does not answer how you’d create the PDF. Whether that happens outside paperless inside paperless does not make a difference. In the end, I need to create a PDF/A out of some images and the question on how to encode these images still remains.