Beyond PRNU: Learning Robust Device-Specific Fingerprint for Source Camera Identification

Area of Science:

• Digital Image Forensics
• Computer Vision
• Signal Processing

Background:

• Source-camera identification is crucial for image forensics.
• Photo Response Non-Uniformity (PRNU) is a common but fragile method, susceptible to manipulations and requiring spatial synchronization.
• Current deep learning models identify camera models but not individual devices and struggle with robustness.

Purpose of the Study:

• To introduce a novel, robust, data-driven, device-specific fingerprint for individual camera identification.
• To overcome the limitations of PRNU and existing deep learning approaches in digital forensics.

Main Methods:

• Extraction of a new device fingerprint from low- and mid-frequency bands of digital images.
• Demonstration of the fingerprint's location-independent and stochastic nature.
• Experimental validation on diverse datasets.

Main Results:

• The new fingerprint successfully identifies individual cameras of the same model.
• It is resilient to spatial synchronization issues, unlike PRNU.
• The fingerprint exhibits high robustness against image manipulations like rotation, gamma correction, and JPEG compression.

Conclusions:

• A novel device-specific fingerprint offers a more robust and reliable solution for source-camera identification in forensics.
• This method addresses key vulnerabilities of PRNU and current deep learning techniques.
• The discovered fingerprint is suitable for practical forensic scenarios demanding high accuracy and resilience.