Source Camera Forensics (SCF) techniques play a crucial role in digital investigations, particularly for attributing the origin of incriminating content. While traditional SCF methods, such as Sensor Pattern Noise (SPN), perform well in verification tasks, they often fall short for large-scale screening scenarios. This article proposes a shift in evaluation focus, from verification-oriented metrics (e.g., False-Positive Rate), which aim to prevent false convictions, to investigation-oriented metrics (e.g., Recall), which prioritize minimizing evidence loss. To this end, we evaluate three approaches: the classic SPN method; CompaRe, an efficient SPN-based variant; and the Media Source Similarity Hash (MSSH), a non-SPN approach that leverages JPEG structural metadata. On a contemporary dataset, MSSH achieves perfect Recall (1.0), albeit with a lower Precision (0.25). In contrast, the classic SPN and the CompaRe approach reach higher Precision values (up to 0.6 and 0.7, respectively), but their Precision drops below 0.1 for Recall values exceeding 0.7, rendering them unsuitable for the given use case in which evidence preservation is critical. Additionally, MSSH offers a speedup of over 500× compared to SPN-based methods and demonstrates its suitability for large-scale investigations.      «

Source Camera Forensics (SCF) techniques play a crucial role in digital investigations, particularly for attributing the origin of incriminating content. While traditional SCF methods, such as Sensor Pattern Noise (SPN), perform well in verification tasks, they often fall short for large-scale screening scenarios. This article proposes a shift in evaluation focus, from verification-oriented metrics (e.g., False-Positive Rate), which aim to prevent false convictions, to investigation-oriented met...     »