PRNU-Bench: A Novel Benchmark and Model for PRNU-Based Camera Identification

2025-09-22

AI Summary

To address the challenge of camera identification based on photo-response non-uniformity (PRNU) in real-world scenarios—particularly the insufficient robustness evaluation under uncontrolled “in-the-wild” conditions—this paper introduces the first large-scale PRNU-based camera identification benchmark, comprising over 120 devices and 13,000 real-scene images. Methodologically, we depart from conventional contrastive learning paradigms and propose an end-to-end denoising autoencoder–convolutional network hybrid architecture, which takes the Hadamard product of reference and query PRNU signals as input to perform 1:N camera verification. This fusion mechanism explicitly enhances the discriminability of PRNU features. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art methods relying on denoising autoencoders or contrastive learning, achieving substantial gains in both identification accuracy and cross-device generalization under realistic imaging conditions.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Large-scale security measurements
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

We propose a novel benchmark for camera identification via Photo Response Non-Uniformity (PRNU) estimation. The benchmark comprises 13K photos taken with 120+ cameras, where the training and test photos are taken in different scenarios, enabling ``in-the-wild'' evaluation. In addition, we propose a novel PRNU-based camera identification model that employs a hybrid architecture, comprising a denoising autoencoder to estimate the PRNU signal and a convolutional network that can perform 1:N verification of camera devices. Instead of using a conventional approach based on contrastive learning, our method takes the Hadamard product between reference and query PRNU signals as input. This novel design leads to significantly better results compared with state-of-the-art models based on denoising autoencoders and contrastive learning. We release our dataset and code at: https://github.com/CroitoruAlin/PRNU-Bench.

Problem

• Proposes a benchmark for camera identification using PRNU estimation
• Introduces a hybrid model for 1:N camera verification from PRNU signals
• Addresses limitations of contrastive learning with a Hadamard product input

Innovation

• Hybrid architecture with denoising autoencoder and convolutional network
• Uses Hadamard product between reference and query signals
• Novel benchmark with in-the-wild evaluation scenarios