IMAGE CLOAKING AND PROCESSING

COMPUTER SCIENCE AND ENGINEERING

Today’s proliferation of powerful facial recognition systems poses a real threat to personal privacy. Ubiquitous facial recognition is a serious threat to privacy. The idea that the photos we share are being collected by companies to train algorithms that are sold commercially is worrying. Anyone can buy these tools, snap a photo of a stranger, and find out who they are in seconds. Regulations can and will help restrict the use of machine learning by public companies but will have negligible impact on private organisations, individuals, or even other nation states with similar goals. So, how do we protect ourselves against unauthorized third parties building facial recognition models that recognize us wherever we may go? This software uses Artificial Intelligence to subtly and almost imperceptibly alter one’s photos in order to trick facial recognition systems. The way the software works is a bit complex. Running one’s photos through this application does not make one invisible to any facial recognition exactly. Instead, the software makes subtle changes to one’s photos so that any algorithm scanning those images in future sees one as a different person altogether. Essentially, cloaking on one’s photos is like adding an invisible mask or filter to one’s photos. This software creates an inaccurate image without significantly distorting the photo or by adding conspicuous patches or filters, which is used against unauthorised facial recognition models. This is achieved by adding imperceptible pixel-level changes, ‘cloaks’, to the image. For example, a user who wants to share content, specifically facial image, on social media or the public web, can add small, imperceptible alterations to their photos before uploading them. If collected by a third-party tracker and used to train a facial recognition model to recognise the user, these ‘cloaked images’ would produce functional models that consistently misidentify that user.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.