Rethinking Image Sensor Noise for Forensic Advantage

TL;DR

This paper reexamines sensor pattern noise in images, revealing additional noise sources beyond traditional models, and proposes a unified approach to improve forensic image provenance analysis with fewer resources.

Contribution

It introduces a comprehensive noise analysis that incorporates multiple noise sources, enhancing fingerprint extraction methods for forensic purposes.

Findings

Additional noise sources can be identified within the noise residue.

A unified noise model improves fingerprinting efficiency.

Fewer resources are needed for comparable forensic performance.

Abstract

Sensor pattern noise has been found to be a reliable tool for providing information relating to the provenance of an image. Conventionally sensor pattern noise is modelled as a mutual interaction of pixel non-uniformity noise and dark current. By using a wavelet denoising filter it is possible to isolate a unique signal within a sensor caused by the way the silicon reacts non-uniformly to light. This signal is often referred to as a fingerprint. To obtain the estimate of this photo response non-uniformity multiple sample images are averaged and filtered to derive a noise residue. This process and model, while useful at providing insight into an images provenance, fails to take into account additional sources of noise that are obtained during this process. These other sources of noise include digital processing artefacts collectively known as camera noise, image compression artefacts, lens artefacts, and image content. By analysing the diversity of sources of noise remaining within the noise residue, we show that further insight is possible within a unified sensor pattern noise concept which opens the field to approaches for obtaining fingerprints utilising fewer resources with comparable performance to existing methods.