A Critical Visual Analysis of Hardware Fingerprints in Low-Light RAW Video

Digital image sensors operating under extreme lowlight conditions produce imagery that is dominated not only by stochastic photon shot noise but also by a persistent, spatially structured layer of deterministic bias known as Fixed-Pattern Noise (FPN). This paper presents an empirical, graphical analysis of the stability, magnitude, and structural origins of FPN across 14 commercial smartphone camera modules, utilising the AIM 2025 Low-Light RAW Video dataset captured at approximately 1 lux. By extracting and systematically analysing FPN residual heatmaps, spatial Power Spectral Density (PSD) profiles, and hierarchical clustering dendrograms across five physically distinct scenes, the following principal findings are established. First, FPN constitutes a stable, scene-independent spatial pattern whose structure is governed primarily by the sensor hardware. Second, sharp harmonic spikes in row-wise PSD profiles implicate periodic column-parallel readout electronics as the dominant source of deterministic banding. Third, the RMS magnitude of FPN varies by a factor of approximately 22× between modules housed within the same handset. Fourth, hierarchical clustering reveals that FPN structural similarity does not conform to brand level grouping; cross-brand affinities instead suggest that the underlying sensor die and module type are more influential than manufacturer identity. These observations highlight the importance of identifying and subtracting deterministic hardware signatures as a prerequisite to effective low-light image enhancement.