#depixelization — Public Fediverse posts

🛠️ Tool
===================

Opening: Depix is a proof-of-concept tool that attempts to recover plaintext from pixelized screenshots produced by a linear box filter. The project targets screenshots where each pixel block was processed independently and averaged, enabling block-wise comparison against a prepared search image.

Key Features:
• Block-matching approach: Depix generates pixelated variants of a provided search image and compares blocks to find one-to-one matches.
• Search-image technique: Use of a De Bruijn sequence in the search image to enumerate possible character sequences in a compact screenshot.
• Color-space handling: Support for gamma-encoded averaging (commonly produced by tools like Greenshot) and an alternative linear sRGB averaging mode.

Technical Implementation:
• The algorithm assumes a linear box filter applied per block and relies on exact block boundaries for matching.
• Depix pixelizes the supplied search image in the same way and performs block-wise comparisons to find matching candidate blocks.
• The tool includes a rudimentary box detector to locate pixel blocks, but detection quality directly affects recovery success.

Use Cases:
• Recovering short secrets exposed in screenshots where only small regions were pixelized and the pixelization method matches the filter model.
• Demonstrating risks of naive pixelization for redaction and guiding defensive choices on stronger redaction methods.

Limitations:
• Success depends on accurate extraction of pixel blocks and on the pixelization being a linear box filter; non-linear or blended filters reduce effectiveness.
• The box detector is brittle and requires well-cropped input; false positives/merged boxes degrade results.
• The approach requires a crafted search image with expected characters and matching font/size/colors to maximize block matches.

References:
• De Bruijn sequence (used to build search images) • linear box filter (assumed pixelization model)

🔹 tool #depixelization #image_processing

🔗 Source: https://github.com/spipm/Depixelization_poc/