Recognised text

Images and PDFs have a Text tab in the file window. There Moliri recognises the text of a scan or a photographed document, shows it next to the image and lets you correct it line by line. The text is read in the browser, with the desktop app Transcribe on your computer, by the AI service or on the server of your instance. Uncertain passages are marked; what the machine reads is checked by a person.

Recognising text

Without recognised text, the tab shows the image and the settings for recognition. The Recognise text button is visible to anyone who may edit the file.

Text tab of a file without recognised text: on the left the image, on the right under Recognised text the fields Place set to With the AI service, Language and Layout, below them the Recognise text button

Setting	Options
Place	In the browser, With Transcribe on this computer, With the AI service, On the server
Language	German, German and English, English, German + Fraktur
Layout	Page (automatic), Lines (index card, form), One block (label), Scattered (writing on objects)

At the start, Place is set to the default place from the profile, page Text recognition . Only the places that can read at the moment are offered: With Transcribe on this computer if the app is running and allows Moliri, With the AI service if the instance has set it up for text recognition, On the server if the service answers. If the default place cannot read at the moment, the place is set to In the browser. If only the browser can read, the field is not shown.

Place	Languages	The image
In the browser	German, German and English, English	stays on the computer
With Transcribe on this computer	all, including German + Fraktur	stays on the computer
With the AI service	all, including German + Fraktur	goes through the server to the service of the AI account
On the server	all, including German + Fraktur	is read on the server

German + Fraktur reads blackletter and roman type on the same page. Language offers only what the chosen place reads: if you switch from Transcribe to the browser, German is read instead of German + Fraktur. The Layout tells the recognition how the text sits on the image: a book page, an index card with fields and lines, a single label, or inscriptions scattered across an object.

In the browser and with Transcribe, Moliri reads straight away. The tab shows Recognising the text in the browser … with the progress in per cent, or Transcribe is recognising the text …. Leave the window open until the text appears.

On the server the job waits or runs; the tab shows its state and, if the service is busy, the position in the queue and how long it has been waiting. Text recognition runs on the server with low priority. The text appears even if the window is closed. For a multi-page PDF, recognition reads the first page.

With the AI service, recognition also runs as a job, and the text appears even if the window is closed. While it reads, the tab shows The AI service is recognising the text …. A model on your own server can take several minutes for one page.

The AI service

The AI service is an image model that the instance addresses through its AI account , on a server of its own or at a provider. It also reads handwriting in Latin script: letters, notes and labels in the cursive hand written today. It does not read Kurrent reliably. There the models write fluent sentences that have little to do with the sheet and still look right. The text recognition on the server reads Fraktur more accurately and faster.

This place is only available where the AI account has a Model for text recognition and the instance has not switched off text recognition by the AI service, see Rules and night window . The image goes to the service of the AI account for this. If the model runs on your own server, the image stays in house.

Server busy

If a job waits on the server for longer than about 30 seconds, the tab says that the server is busy at the moment and offers the other places: Recognise in the browser and, if Transcribe is running, Recognise with Transcribe. A click reads the image there straight away, with the same layout and the same language, as far as that place reads it. The job on the server is then dropped. As with Recognise again, corrections stay with their line. A job for the AI service does not get this notice.

Notice in the Text tab: the server is busy right now and another place is faster now, below it the buttons Recognise in the browser and Recognise with Transcribe

Anyone who may only read the file sees only the note that the text appears even if the window is closed.

Images also get their text through a rule for a file keyword, see Rules and night window .

Image and text

The image is on the left, with Zoom in, Zoom out and Home. On it lie the word boxes of the recognised text, uncertain words with dashed outlines; a button hides and shows them. On the right is the text in lines and paragraphs, with uncertain words underlined with dots.

Image and text are linked. A click on a line highlights its boxes in the image and brings them into view, enlarged if they are too small to read. A click on a box selects the line in the text.

The AI service provides no boxes for the individual words and only the approximate position of the lines. Where word boxes are missing, the image marks the whole line. The AI service does not report uncertain words, so its text has no dotted passages. You still have to read it.

Correcting

Anyone who may edit the file can correct it.

Key	Effect
Click in a line	Typing changes the line
`Enter`	To the next line
`↓` at the end of a line	To the next line
`↑` at the start of a line	To the previous line
`Esc`	Takes back what you typed
`Ctrl` + `Z`, `Ctrl` + `Y`	Undo, redo

The line is saved when you leave it. A corrected line has a bar in the margin and can be reset to what was recognised (Back to …); a line you empty drops out of the text. The history of the file shows one entry Recognised text corrected per session, with the text before and after. If someone else has corrected the line in the meantime, the tab loads their version.

Tool	Effect
Search the text	Shows the number of matches and jumps to the previous and next one
n uncertain	Jumps to the next uncertain line that has not been corrected yet
Undo, redo	Like `Ctrl` + `Z` and `Ctrl` + `Y`
Copy	The whole text to the clipboard
Download as a text file	The text with all corrections
Recognise again	Recognises the text once more, for example at a different place or with a different language or layout

Recognise again opens the same settings as Recognise text, with Place, Language and Layout. The corrections stay wherever the line is found again, whichever place read it. The tab says how many corrections no longer have a matching line; they stay saved. If the new recognition fails, the previous text stays.

Every recognition appears in the history of the file as Text recognised (machine), with the recognition before and after and the language. The name of the recognition shows the place, for example browser:tesseract.js-7.0.0 for the browser, desktop:tesseract-5.0.0 for Transcribe, ai: with the name of the model for the AI service and server:tesseract-5.3.0 for the server.

OCR import

The OCR import (Documents and scans) reads at the same places. Its field Text recognition offers In the browser, With Transcribe on this computer, With the AI service and On the server, as far as they can read at the moment, and at the start it is set to the default place from the profile. The languages follow the place as in the Text tab. With the AI service, the documents go through the server to an image model of the instance’s AI account; it reads handwriting too. The import then reads page by page while the dialog is open. PDFs with their own text layer are read by the import directly, without text recognition.

The import checks the pages it has read in the same editor: the sheet with zoom and word boxes next to the text, editing with Enter, arrow keys and Esc, search and jumping to uncertain passages. There each section also belongs to a field.

First step of the OCR import Documents and scans: the drop area for the files, below it the fields Text recognition set to With the AI service, Language, Layout, Objects and Several pictures per object, and the note that documents are not uploaded in the browser and with Transcribe and that with the AI service they go through the server to the instance's AI account