[FEATURE] OCR and Full-Text-Search

[user12257]

Problem Statement

Currently, it is not possible to search for Text that is inside PDFs and images.

Proposed Solution

Text from PDFs and images would be needed to be extracted and stored somewhere for convenient retrieval.

If I am not mistaken, Nextcloud uses Elasticsearch, which offers an experimental rust client and could be used for that here as well.

For extracting the text from PDFs, I think that should be either doable by, just extracting the text from a computer generated PDF with readable text, or by using ocr.

On the first look, this project, looks like it could be helpful. Although it cannot extract text directly from PDF, converting PDF to PNG first could be a workaround.

User Impact

The user could search for keywords and find all documents it appears in, as well as the page and location.

An important question would be, how to structure the data, and how to efficiently store, retrieve and display it.

I could take a deeper look into actually implementing this, though time wise, it would probably be in a while.

[EdouardVanbelle]

Hello interesting feature,

I am questioning how we can have a such feature but from client side

(example : client app parse document during upload + enrich itself a database that can be on server side, client app deals with this database when doing lookup)

Because when we will implement the end to end encryption, 2 options:

• either client app can retrieve it's database and search in it
• or no lookup possible when fully encrypted

[user12257]

Good point, I actually did not think of E2EE.

I think, from what I can find, in Nextcloud the Full text search does not work in combination with E2EE.

To me that sounds like a reasonable tradeoff.

On clients, with file sync some local tool might be helpful, like ripgrep.