The Unredacter toolkit is introduced, allowing you to restore the original text after hiding it using filters based on pixelation. For example, the program can be used to identify sensitive data and passwords pixelated in screenshots or snapshots of documents. The algorithm implemented in Unredacter is claimed to be superior to previously available similar utilities such as Depix, and was successfully used to pass the pixelated text detection test proposed by the Jumpsec lab. The program code is written in TypeScript and distributed under the GPLv3 license.
To restore text in Unredacter, the reverse selection method is used, according to which a part of the original pixelated image is compared with a variant synthesized by enumerating pairs of characters pixelated with different shifts and changed characteristics. During the enumeration, the variant that most closely matches the original fragment is gradually selected. For successful work, it is required to correctly guess the size, type and parameters of font indents, as well as calculate the cell size in the pixelation grid and the position of this grid overlay on the text (grid offset options are sorted automatically).
Additionally, the DepixHMM project can be noted, within the framework of which a version of the Depix utility was prepared, translated into an algorithm based on a hidden Markov model, thanks to which it was possible to achieve an increase in the accuracy of character reconstruction.
Source: opennet.ru