Pose-Based Sign Language Appearance Transfer

TL;DR

This paper presents a pose-based appearance transfer method for sign language that enhances privacy by obfuscating signer identity while maintaining sign content, with tradeoffs in recognition accuracy.

Contribution

It introduces a novel appearance transfer technique for sign language skeletal poses that balances privacy preservation with sign recognition performance.

Findings

Reduces signer identification accuracy

Slightly decreases sign recognition performance

Improves pose-based rendering and sign stitching

Abstract

We introduce a method for transferring the signer's appearance in sign language skeletal poses while preserving the sign content. Using estimated poses, we transfer the appearance of one signer to another, maintaining natural movements and transitions. This approach improves pose-based rendering and sign stitching while obfuscating identity. Our experiments show that while the method reduces signer identification accuracy, it slightly harms sign recognition performance, highlighting a tradeoff between privacy and utility. Our code is available at https://github.com/sign-language-processing/pose-anonymization.