SpecGuard: Spectral Projection-based Advanced Invisible Watermarking

TL;DR

SpecGuard addresses the challenge of robust, invisible image watermarking by embedding in the high-frequency spectral domain via wavelet projection and FFT-based spectral projection. The encoder refines the embedding with multiple convolutions and a radial mask controlled by radius $r$ and strength $s$, while the decoder uses a learnable threshold $\theta$ guided by Parseval-based energy considerations to reliably recover the watermark under diverse distortions. A joint loss $L = \lambda_{\text{enc}} L_{\text{enc}} + \lambda_{\text{dec}} L_{\text{dec}}$ optimizes invisibility and recoverability, with theoretical support including a proof for the $S_{HH}$ band and a maximum capacity formula $C_{\max}$ in terms of image size, wavelet level, and spectral fraction. Empirical results on multiple datasets demonstrate high invisibility (PSNR/SSIM), high capacity across bit lengths, and strong robustness against distortions, regeneration, and adversarial attacks, outperforming state-of-the-art watermarking methods. The work offers practical impact through strong security guarantees and reproducibility via released code.

Abstract

Watermarking embeds imperceptible patterns into images for authenticity verification. However, existing methods often lack robustness against various transformations primarily including distortions, image regeneration, and adversarial perturbation, creating real-world challenges. In this work, we introduce SpecGuard, a novel watermarking approach for robust and invisible image watermarking. Unlike prior approaches, we embed the message inside hidden convolution layers by converting from the spatial domain to the frequency domain using spectral projection of a higher frequency band that is decomposed by wavelet projection. Spectral projection employs Fast Fourier Transform approximation to transform spatial data into the frequency domain efficiently. In the encoding phase, a strength factor enhances resilience against diverse attacks, including adversarial, geometric, and regeneration-based distortions, ensuring the preservation of copyrighted information. Meanwhile, the decoder leverages Parseval's theorem to effectively learn and extract the watermark pattern, enabling accurate retrieval under challenging transformations. We evaluate the proposed SpecGuard based on the embedded watermark's invisibility, capacity, and robustness. Comprehensive experiments demonstrate the proposed SpecGuard outperforms the state-of-the-art models. To ensure reproducibility, the full code is released on \href{https://github.com/inzamamulDU/SpecGuard_ICCV_2025}{\textcolor{blue}{\textbf{GitHub}}}.

Figures (10)

• Figure 1: Image authentication using our proposed SpecGuard.
• Figure 2: Architecture of the proposed SpecGuard watermarking method involves encoding a binary message $M$ into the high-frequency band of the cover image $I$ using wavelet and spectral projection and learning to decode the embedded message.
• Figure 3: Some best results for cover vs watermarked images with PSNR/SSIM ($\uparrow$) scores showing minimal visual degradation when watermarked using proposed SpecGuard.
• Figure 4: Robustness validation of our proposed SpecGuard under different distortion attacks, including geometric transformations: Geo (rotation, cropping), photometric modifications (brightness, contrast), and degradations: Deg (blur, noise, JPEG compression).
• Figure 5: Comparison of SOTA watermarking methods in terms of average TPR@0.1%FPR (90% of watermarked images are correctly detected at 0.1% false positive rate) under different attacks.