The Adobe Hidden Feature and its Impact on Sensor Attribution

TL;DR

This work identifies a periodic 128×128 watermark embedded by Adobe Lightroom/Camera Raw before 8‑bit quantization, which creates false positives in PRNU-based sensor attribution. The authors diagnose the leak by linking FP sources to Adobe software EXIF data and demonstrating a reproducible watermark in residuals across 16‑to‑8‑bit development; they show the watermark is content- and architecture-dependent and not present in 16‑bit TIFF exports. They propose two removal strategies—residual-domain cancellation and image-domain cancellation—alongside JPEG dimples mitigation, both of which effectively suppress FP in a Leica Q2 case study. The findings have practical forensic implications, enabling more reliable sensor attribution when dealing with Adobe-developed images, while highlighting the need for careful analysis of watermark-like patterns in forensic pipelines. The work also discusses the broader interpretation of dithering as a potential motive for watermark embedding and suggests avenues for further investigation into non-Adobe FP sources.

Abstract

If the extraction of sensor fingerprints represents nowadays an important forensic tool for sensor attribution, it has been shown recently that images coming from several sensors were more prone to generate False Positives (FP) by presenting a common "leak". In this paper, we investigate the possible cause of this leak and after inspecting the EXIF metadata of the sources causing FP, we found out that they were related to the Adobe Lightroom or Photoshop softwares. The cross-correlation between residuals on images presenting FP reveals periodic peaks showing the presence of a periodic pattern. By developing our own images with Adobe Lightroom we are able to show that all developments from raw images (or 16 bits per channel coded) to 8 bits-coded images also embed a periodic 128x128 pattern very similar to a watermark. However, we also show that the watermark depends on both the content and the architecture used to develop the image. The rest of the paper presents two different ways of removing this watermark, one by removing it from the image noise component, and the other by removing it in the pixel domain. We show that for a camera presenting FP, we were able to prevent the False Positives. A discussion with Adobe representatives informed us that the company decided to add this pattern in order to induce dithering.

Figures (8)

• Figure 1: The original $\mathrm{PCE}$s for the Leica Q2 camera and 7 different devices. The first 6 labels are the FlickR identification names and the last one is images coming from a Q2 camera shared by the Leica Shop in Lille and developed using Adobe Lightroom in 8-bit Tiff format. Matching and mismatching tests are reported in green and red, respectively. The threshold of 60 is highlighted by the red dashed line.
• Figure 2: Crop of the autocorrelation function of one residual associated with a source generating FP.
• Figure 3: Average of non-overlapping $128\times 128$ patches of an image residual.
• Figure 4: Changes introduced after 16bit $\rightarrow$ 8bit quantization of two constant TIFF images of size $128\times 128$. (a): 382 (16bit) $\rightarrow$ 1 (8bit), (b): 638 (16bit) $\rightarrow$ 2 (8bit). (c): Their difference image.
• Figure 5: For a constant RAW image with added Gaussian white noise with $\sigma=2$, developed with Lightroom to JPEG, we show (a) the estimated watermark when developed to QF 100 on Apple M1 chip, (b) the difference between estimated watermarks at QF 100 when the same image was developed with Apple M1 chip and Intel chip, (c) estimated watermark when developed to QF 100 on Apple M1 chip.