The Unredacter tool has been introduced, allowing for the recovery of original text after it has been hidden using pixelation-based filters. For instance, the program can be used to identify sensitive data and passwords that have been pixelated in screenshots or document images. It is claimed that the algorithm implemented in Unredacter surpasses similar utilities previously available, such as Depix, and has even been successfully used to pass a test for detecting pixelated text proposed by the Jumpsec laboratory. The program's code is written in TypeScript and is distributed under the GPLv3 license.
To recover text in Unredacter, a reverse matching method is employed, which involves comparing part of the original pixelated image with a variant synthesized through the brute-force pairing of pixelated character pairs with various shifts and altered characteristics. During the iterations, a variant that closely matches the original fragment is gradually selected. For successful operation, it is crucial to correctly guess the size, type, and parameters of the font's indentations, as well as to calculate the cell size in the pixelation grid and the overlay position of this grid on the text (the grid's offset variants are automatically iterated).
Additionally, the DepixHMM project can be noted, within which a variant of the Depix utility has been prepared, translated to an algorithm based on a hidden Markov model, which has achieved improved accuracy in recreating characters.
Source: opennet.ru
