OCR German: searchable PDFs with umlauts and ß

German OCR goes wrong in a few predictable places: an umlaut read as a plain vowel, ß read as B, a case number with one digit off. Inklight reads German scans on your own Windows or Linux PC and writes a searchable PDF beside each one. German is one of the six languages its recognition is tested on, and a German word list ships with the app. Inklight outlines the words it is unsure about, and you check them.

How it works in Inklight

Drop the scans or PDFs on the Processing Queue, or scan straight from your scanner. Each file becomes a Job. Inklight recognizes it in the background and writes {name}_ocr.pdf next to the original. You do not pick a language. Inklight reads each document against the word lists of the languages it finds on the page, up to two, so a German letter with an English paragraph needs no setting.

When a file is done, its card says how many words are left to check. Click Check words to step through them, riskiest first. A German court file number such as 4711/24 counts as an identifier and comes before ordinary words:

A word read without its umlaut is not in the German word list, so it does not slip through as a known word. When the two recognition models read a word differently, both readings appear and you press 1 or 2:

Enter accepts a word, typing replaces it, Tab skips it. Inklight writes every decision into the searchable PDF’s text layer straight away, so a search for “Gebühr” finds the page.

The Check pass also groups the same word across files. If Gebuhr turns up on three pages of two tax notices, you see it once, marked “×3 in 2 documents”, with the line of each other copy below it. One decision writes Gebühr into every copy whose box stays ticked. Clear a box to decide that copy on its own.

What you get

• Searchable PDFs from German scans, PDFs and photos, made on your machine. Nothing is uploaded. Every plan.
• ä, ö, ü and ß kept in the text layer as read, so search and copy work with the real spelling.
• German PDFs that already have text on every page are left alone. In a PDF where only some pages have text, Inklight recognizes the others and keeps the rest of the file. Every plan.
• The shipped German word list, used to decide which uncertain words need a look. Every plan.
• IBAN check: an IBAN whose checksum fails is the first word the Check pass shows. Every plan.
• Aktenzeichen ranked as identifiers, ahead of ordinary words. Every plan.
• Output named after the Aktenzeichen: put {aktenzeichen} in the name pattern and 1234/24 becomes 1234_24 in the file name. Business.
• Clean pages without the footer. Free writes a small footer on every page; Personal and Business do not.

Limits

• Inklight reads printed Latin-script text. Handwriting is not read.
• The six tested languages are English, German, French, Spanish, Italian and Dutch. Old blackletter print (Fraktur) is not among them.
• Word lists ship for 13 languages, German, English and Dutch among them. There is no Latin list, so no list can clear an uncertain word in a phrase such as de lege ferenda, and it goes to the Check pass.
• Only words the Check pass flags are outlined. A word misread with high confidence by both models is not, so compare amounts and file numbers you rely on with the scan. Check and fix OCR errors shows how.
• Windows and Linux only. There is no macOS version.

More guides

• OCR French: accents and œ
• OCR Spanish: accents and ñ
• ABBYY FineReader alternative: buy once, run on Linux

Get started

Download Inklight for Windows or Linux. Every install starts with a seven-day Business trial; see the plans for what stays free.