Finding of a handwritten encrypted document, without associated decryption, and without a known key.
The user uploads a photograph or scan of the document to the software, called Glyphea.
Alongside the document, the user provides information on the period, country of origin and presumed or confirmed language of the encrypted text, if possible.
The transliteration module recognizes handwritten characters individually and converts the uploaded documents into text files that can be read by a computer.
To do so, we use neural networks that have been specially trained for this task. Therefore, the dedicated artificial intelligence only performs a transliteration task (recognizing and naming each character) and not a deciphering task (restoring the content).
The text that has been transliterated is now ready for cryptanalysis.
The transliterated text is directed to the most appropriate cryptanalysis algorithm depending on the metadata provided (location, date and language). The attack method used depends on the cryptographic system employed, which is often deducible from the document's date and location.
The user is provided with an initial transcription, and can then indicate or correct any inconsistencies and complete any remaining undeciphered characters (such as names and abbreviations).
The objective has been achieved: the user obtains a consolidated version of the decrypted text...
... and the key. It usually takes the form of a clear/cipher correspondence table.
