Fragile Watermarking for Image Certification Using Deep Steganographic Embedding

TL;DR

This paper tackles the integrity of ICAO-compliant facial images by introducing a fragile watermarking scheme built on deep steganographic embedding. It embeds a known marker into the official photo at issuance, so any post-issuance modification degrades the recovered marker, enabling tamper detection and forensic analysis. The authors evaluate two embedding models, Stegformer and SteGuz, across a suite of manipulations (compression, resizing, noise, blur, sharpening, morphing) and show high recoverability of the marker and strong manipulation classification performance, including cross-model generalization. They demonstrate that deep steganography-based fragile watermarking can provide actionable forensic signals for biometric document integrity and discuss practical deployment implications and future directions.

Abstract

Modern identity verification systems increasingly rely on facial images embedded in biometric documents such as electronic passports. To ensure global interoperability and security, these images must comply with strict standards defined by the International Civil Aviation Organization (ICAO), which specify acquisition, quality, and format requirements. However, once issued, these images may undergo unintentional degradations (e.g., compression, resizing) or malicious manipulations (e.g., morphing) and deceive facial recognition systems. In this study, we explore fragile watermarking, based on deep steganographic embedding as a proactive mechanism to certify the authenticity of ICAO-compliant facial images. By embedding a hidden image within the official photo at the time of issuance, we establish an integrity marker that becomes sensitive to any post-issuance modification. We assess how a range of image manipulations affects the recovered hidden image and show that degradation artifacts can serve as robust forensic cues. Furthermore, we propose a classification framework that analyzes the revealed content to detect and categorize the type of manipulation applied. Our experiments demonstrate high detection accuracy, including cross-method scenarios with multiple deep steganography-based models. These findings support the viability of fragile watermarking via steganographic embedding as a valuable tool for biometric document integrity verification.

Figures (6)

• Figure 1: Overview of the proposed methodology, structured into three main phases: embedding, alteration and classification.
• Figure 2: Overview of the steganographic certification process and its impact on image quality: a) original input image; b) original secret image to be embedded; c) stego image generated using Steguz; d) secret image recovered from the Steguz stego image; e) stego image generated using Stegformer; f) secret image recovered from the Stegformer stego image.
• Figure 3: Recovered secret images examples using Steguz after applying manipulations: a) JPEG compression ($Q_{F}=80$), b) Gaussian blur ($K_{G}=7$), c) Gaussian noise ($\sigma=8$), d) resize ($R_{F}=85\%$), e) salt & paper noise ($P_{SP}=\left(0.3, 0.01\right)$), f) sharpening ($S_{F}=0.5$), g) morphing ($\alpha_{M}=0.9$).
• Figure 4: Recovered secret images examples using Stegformer after applying manipulations: a) JPEG compression ($Q_{F}=80$), b) Gaussian blur ($K_{G}=7$), c) Gaussian noise ($\sigma=8$), d) resize ($R_{F}=85\%$), e) salt & paper noise ($P_{SP}=\left(0.3, 0.01\right)$), f) sharpening ($S_{F}=0.5$), g) morphing ($\alpha_{M}=0.9$).
• Figure 5: Effect of transformations on image recovery. Error bars reflect variation within each manipulation type.