Web Photo Source Identification Based on Neural Enhanced Camera Fingerprint

The authors address the challenge of tracing the source device of online photos for copyright protection, forensic evidence, and authentication by proposing a systematic approach that integrates camera fingerprint extraction with cryptographic techniques.

Overall Solution : A three‑stage pipeline (registration, generation, verification) is designed, where each captured photo undergoes fingerprint extraction and is linked to a proof or signature, enabling verification while preserving privacy.

Camera Fingerprint Extraction Network : A novel neural network architecture is introduced to extract camera fingerprints from RAW images, achieving a reduction of identification error from 40.62% to 2.345% on public test sets, outperforming traditional PRNU and denoising networks.

Fingerprint Optimization Modules : Three modules—Spatially Splitting, Block Filtering, and Burst Integration—are added to improve robustness and security by separating private/public parts, weighting block brightness, and aggregating burst shots.

Cryptographic Schemes : Two schemes are proposed: (1) an authorization scheme using fuzzy extraction and digital signatures based on random projection and polar codes, and (2) a zero‑knowledge proof scheme that encodes the fingerprint extraction and matching process into a ZKP circuit, providing rich proof semantics for auditors.

Experimental Results : Experiments on a dataset of 150,000 RAW iPhone photos (72 cameras) show the proposed network surpasses traditional methods in both identification accuracy and confusion‑matrix scores; ablation studies confirm each optimization module contributes positively.

Conclusion : By combining AI‑driven fingerprint extraction with strong cryptographic protocols, the system offers a reliable, privacy‑preserving solution for photo source identification, raising the cost and difficulty of device‑owner forgery.