SpreadMark: Robust Image Watermarking via Spread-Spectrum Embedding

Authors: Wei Song, Yuxin Cao, Zhenchang Xing, Liming Zhu, Jin Song Dong, Yulei Sui, Jingling Xue

Published: 2026-08-04 05:52:17+00:00

Comment: 12 pages, 6 figures

AI Summary

SpreadMark introduces a robust image watermarking technique that utilizes spread-spectrum embedding within a neural post-hoc architecture. By spreading each message bit as a dense pseudo-random codeword across the entire image, the method significantly enhances robustness against deliberate removal attacks like regeneration and latent-space sparsification, while maintaining imperceptibility.

Abstract

Invisible image watermarks are increasingly used for deepfake detection and provenance tracking, where they must survive not only incidental distortions but also deliberate removal. We revisit spread-spectrum embedding, a classical watermarking principle, inside a modern neural post-hoc watermarking architecture. Our starting point is a measurement: in existing encoder-decoder schemes each message bit occupies only a small fraction of the image, a shared contributing factor to their fragility, since removal then need only disturb the region a bit occupies. SpreadMark instead spreads each bit as a dense pseudo-random codeword over the whole image and recovers it by matched-filtering a learned cover-suppressed chip representation, with a parallel convolutional decoding path and sparsification-aware training. A conditional chip-space analysis shows that, under a codeword-independent perturbation model, dense spreading increases the budget required to disrupt matched-filter recovery. Evaluated on COCO and DIV2K against nine schemes, SpreadMark is the only evaluated method retaining high detection under both the regeneration and the latent-space sparsification settings we test, with competitive JPEG and additive-noise robustness. It keeps the embedded watermark imperceptible, maintaining high perceptual quality on both COCO and DIV2K.

Key findings

SpreadMark demonstrates superior robustness against regeneration and latent-space sparsification attacks compared to nine other watermarking schemes, being the only evaluated method to retain high detection rates under these conditions. It maintains high perceptual quality (30.1 dB PSNR on COCO) and competitive robustness against JPEG and additive noise, though it shows limitations against geometric distortions (blur, crop) and high-energy adversarial removal at lower embedding gains.

Approach

SpreadMark embeds a watermark by superimposing an explicit spread-spectrum signal onto a learned additive residual, where each message bit is represented by a fixed pseudo-random codeword spanning the entire image. The decoder recovers the watermark by matched-filtering a learned cover-suppressed 'chip' representation against the same codeword bank, complemented by a parallel convolutional decoding path and sparsification-aware training.

Datasets

COCO, DIV2K

Model(s)

UNKNOWN

Author countries

Australia, Singapore

← Previous