Apple's Synthetic Defocus Noise Pattern: Characterization and Forensic Applications

TL;DR

Apple's portrait-mode processing embeds a distinctive SDNP that can mislead PRNU-based attribution. The authors formalize a BP-based model, extract and characterize the base pattern under Natural Light and Stage Light Mono, and estimate ISO- and brightness-dependent scaling functions to recover a robust base pattern. They demonstrate BP-driven forensic applications, including reliable Apple portrait detection, model/version tracing, and a BP-aware PRNU framework that significantly reduces SDNP-induced false positives. The results show improved camera source verification in portrait-mode scenarios and suggest practical paths for robust forensic analysis despite evolving device and OS changes.

Abstract

iPhone portrait-mode images contain a distinctive pattern in out-of-focus regions simulating the bokeh effect, which we term Apple's Synthetic Defocus Noise Pattern (SDNP). If overlooked, this pattern can interfere with blind forensic analyses, especially PRNU-based camera source verification, as noted in earlier works. Since Apple's SDNP remains underexplored, we provide a detailed characterization, proposing a method for its precise estimation, modeling its dependence on scene brightness, ISO settings, and other factors. Leveraging this characterization, we explore forensic applications of the SDNP, including traceability of portrait-mode images across iPhone models and iOS versions in open-set scenarios, assessing its robustness under post-processing. Furthermore, we show that masking SDNP-affected regions in PRNU-based camera source verification significantly reduces false positives, overcoming a critical limitation in camera attribution, and improving state-of-the-art techniques.

Figures (14)

• Figure 1: Values of $\eta$ for the 3 iPhone 11 Pro users with PRNU collisions in IULIANI_2021. From left to right, each user's PRNU fingerprint is tested against the other two. The upper pie charts show the distribution of Reference and Test images per user, indicating the proportion of Portrait and Photo images.
• Figure 2: Apple's portrait mode block diagram highlighting the 3 main stages.
• Figure 3: Block diagram of the extraction process for the averaged residual matrix in each portrait lighting mode under study: $\bar{\mathbf{W}}_{\text{NL}}$ (a) and $\bar{\mathbf{W}}_{\text{SLM}}$ (b).
• Figure 4: Upper-left $128\times128$ patch of $\hat{\mathbf{P}}_{\text{NL}}$ estimated from 12MP HEIF images on the iPhone 15 (a). Autocorrelation of $\hat{\mathbf{P}}_{\text{NL}}$ limited to $k,l\in[-5,5]$ (b).
• Figure 5: Pixel value distribution in $\mathbf{Z}_{\text{SLM}}$ for different ISO values.