Watermarking Images in Self-Supervised Latent Spaces

TL;DR

This paper introduces a self-supervised method for embedding robust watermarks into deep network latent spaces, outperforming previous zero-bit techniques and matching state-of-the-art multi-bit watermarking methods.

Contribution

It presents a novel self-supervised approach to watermarking in latent spaces that is resolution-agnostic and highly robust to various transformations.

Findings

Outperforms previous zero-bit watermarking methods

Achieves performance comparable to end-to-end trained watermarking architectures

Operates effectively across various image resolutions and transformations

Abstract

We revisit watermarking techniques based on pre-trained deep networks, in the light of self-supervised approaches. We present a way to embed both marks and binary messages into their latent spaces, leveraging data augmentation at marking time. Our method can operate at any resolution and creates watermarks robust to a broad range of transformations (rotations, crops, JPEG, contrast, etc). It significantly outperforms the previous zero-bit methods, and its performance on multi-bit watermarking is on par with state-of-the-art encoder-decoder architectures trained end-to-end for watermarking. The code is available at github.com/facebookresearch/ssl_watermarking