AI-driven anonymisation that preserves human performance
Figure 4 - Four pairs of frames from a head-swap anonymisation test. The left image of each pair shows a real frame of a test contributor, while the right image of each pair shows the output of the head-swap pipeline

AI-driven anonymisation that preserves human performance

Tech Papers 2026: This paper presents two AI-driven workflows designed to preserve emotional nuance while ensuring anonymity.

Abstract

The BBC frequently protects contributor identities, but traditional methods of hiding visual appearance, like blurring or actor reconstructions, often distance the audience and alter interview tone. 

The first is a bespoke head-swapping pipeline featuring a Y-shaped autoencoder and super-resolution model.

Latest Technical paper
Favourites:

Registered users only: Login

Share this:
Other themes: