Performance Evaluation of Source Camera Attribution by Using Likelihood Ratio Methods

Area of Science:

• Digital Forensics
• Image Analysis
• Computer Vision

Background:

• Current source camera attribution methods rely on similarity scores lacking probabilistic interpretation.
• Standard evaluation metrics like ROC curves offer limited usability in forensic investigations.
• This hinders informed decision-making for triers of fact.

Purpose of the Study:

• To demonstrate the transition from similarity scores to likelihood ratios for source camera attribution.
• To enhance the probabilistic interpretation and forensic usability of digital evidence evaluation.
• To integrate findings into existing forensic casework.

Main Methods:

• Likelihood ratios were calculated from Photo Response Non-Uniformity (PRNU) source attribution similarity scores.
• Experiments compared different strategies for digital images and videos, accounting for their unique characteristics.
• Results were formatted according to guidelines for validating forensic likelihood ratio methods.

Main Results:

• Successfully transitioned from similarity scores to probabilistic likelihood ratios.
• Demonstrated the utility of likelihood ratios in source camera attribution for digital evidence.
• Provided a framework compatible with forensic validation standards.

Conclusions:

• Likelihood ratios offer a more interpretable and probabilistically sound approach to source camera attribution.
• This transition significantly enhances the value of digital evidence in forensic investigations.
• The proposed method supports more informed decision-making in legal contexts.