Text Depixelization (github.com)

Depix is an open-source proof‑of‑concept tool that can recover plaintext from pixelated screenshots that were created with a linear box (block) filter. Originally built to rescue a partially pixelated password, the repo attracted media attention (briefly private to avoid hype) and demonstrates a straightforward but effective attack on naïve pixelization: by rendering a De Bruijn sequence (all character combinations) with the same editor/font and pixelizing it, Depix searches for block-by-block matches between the unknown screenshot and the generated corpus. Single-match blocks are accepted, multi-match blocks are resolved by geometric consistency with neighbors or by averaging matches; Greenshot (gamma-encoded averaging) and GIMP (linear sRGB averaging) modes are supported. This matters because many people assume pixelization protects secrets in screenshots; Depix shows that simple box-averaging leaks enough per-block information to invert text when fonts, rendering offsets, and no additional compression are known. Limitations include brittle block detection, the need for matching font/screen settings or a good search image, failure under additional compression or sub-pixel text positioning, and integer-aligned block assumptions. The author suggests improvements (more averaging models, filters, HMM-based approaches) and points to related work (DepixHMM) and recent advances applying TensorFlow to moving-image depixelation, underscoring both practical risk and active research avenues to harden or attack pixelization.