|Google Open Sources Image Blurring
|Written by Alex Denham
|Tuesday, 03 January 2023
Google has announced open source versions of some of its "privacy-enhancing technologies", aka PETs. These are used to keep users' information anonymous and protected while still letting Google provide recommendations such as autocorrect suggestions and and the low-down in a restaurant's most popular dishes.
Google has a commitment to make these PET technologies freely available via open source projects, and has announced open source versions of image blurring technology, and a faster version of its Fully Homomorphic Encryption (FHE) Transpiler.
The image blurring option is based on Google's internal project, Magritte, which uses Machine Learning (ML) to detect objects using low computational resources, and applies a blur to those objects automatically, as soon as they appear on screen. The tool can also be used to blur arbitrary objects such as license plates.
Google says the code is particularly useful for video journalists who want to provide increased privacy assurances. By using the open-source code, videographers can save time in blurring objects from a video, instead relying on the ML algorithm to detect and blur all the faces in a video.
Magritte is a MediaPipe-based library that provides processing graphs to reliably detect faces, track their movements in videos, and disguise the person's identity by obfuscating their face.
Processing graphs are built from feature subgraphs that solve sub-problems such as detecting or tracking faces (without recognizing them), determining the area to be redacted, and applying de-identification effect to this area (e.g., blurring, pixelization or sticker redaction).
The second improvement is a faster version of the Fully Homomorphic Encryption (FHE) Transpiler released in open-source format last year.
Google has now introduced new circuit optimizations for the transpiler that result in lower computational cost and time. Google estimates that circuit size has decreased by 50% which improves overall speed.
or email your comment to: firstname.lastname@example.org
|Last Updated ( Tuesday, 03 January 2023 )