When |s |t ever a p|pe/bar (|t's never a p|pe)

[beeeeswax]

I assume this is a Tesseract thing and not a fault with SubtitleEdit - does everybody else have to constantly battle with I's and L's being recognised as a pipe/bar when doing OCR? I have never in my life, ever, seen any subtitle ever that contains the pipe character, so why does it gets used as a guess so often?

I would find others' thoughts on this and any tricks to reduce this very interesting. In my opinion, the pipe symbol should never. ever be used as a guess when trying to OCR and I cannot fathom why it is currently.

Cheers!

[niksedk]

Yes, the pipes are from Tesseract.
Could you supply a .sup file or a few images that have this issue? (you can attach / drag-n-drop .zip files here on Github)

[beeeeswax]

Hi Nik, sorry for the long delay. I've been OCRing again recently and hit a few pipes in The Goonies (UHD), sup attached:
goon.zip

[niksedk]

Thx for the file - love OCR'ing new BD Sups :)

Actually, with latest SE beta which uses Tesseract 5.0.1 I got not pipes, yay.
https://github.com/SubtitleEdit/subtitleedit/releases/download/3.6.4/SubtitleEditBeta.zip
(Did not look into if it is Tesseract or SE replace lists that fixes this)

nOCR also works well with this file.

[beeeeswax]

Great news! I will update to the beta and continue my current OCR mission.

I'm doing my whole UHD collection because when I rip the disc and watch in HDR via Kodi/Plex, the subs are far too bright. Using various player's options to have grey/dark grey subs helps but doesn't go far enough imo.

Anyway, if I come across any particularly bad pipe cases going forward, I'll upload the .sup and give you a nudge here.

[beeeeswax]

Hm, I might need a little more help here, I must have mis-configured something. I've updated to the beta and tried OCRing the Goonies again, I'm still getting pipes. Here's the result of the OCR which shows my settings too:

https://i.imgur.com/xtso5fV.jpg

I thought maybe it was because I have the "Engine mode" set to "Original Tesseract only" (I like italics) so I tried again with "Default" and unticked the "Fallback to Tesseract 3.02" option. This looks to solve the pipe problem but it also makes many basic mistakes now, example:

https://i.imgur.com/5MSRLU9.jpg

Is it just a case of picking my poison or have I done something obviously wrong here?

edit: unticking the [Italic] box has solved Default mode making lots of mistakes, it looks very accurate now albeit without any italics at all. So I guess I have to choose between general accuracy and italic support?

[niksedk]

Works fine here. You probably need the beta dictionary/ocr folder too...