A New Hope for PRNU on Modern Smartphones? Evidence That Device-Specific Fingerprints Survive Computational Photography

Benjamin Loison; Quentin Bammey; Thibaud Ehret; Rafael Grompone von Gioi; Pablo Musé; Marina Gardella

cs.CVOct 18, 2026

A New Hope for PRNU on Modern Smartphones? Evidence That Device-Specific Fingerprints Survive Computational Photography

Authors: Benjamin Loison, Quentin Bammey, Thibaud Ehret, Rafael Grompone von Gioi, Pablo Musé, Marina Gardella

Organizations: Agence Ministérielle pour l'IA de Défense, France · ENS Paris-Saclay, Université Paris-Saclay, CNRS, Centre Borelli, Gif-sur-Yvette, France · Télécom Paris, Institut Polytechnique de Paris, LTCI, Palaiseau, France · IIE, Facultad de Ingeniería, Universidad de la República, Uruguay

Abstract

Photo Response Non-Uniformity (PRNU) is widely regarded as the most reliable marker for source camera identification. Recent studies, however, report high correlations between images and fingerprints from different modern smartphones, which calls into question the discriminative power of PRNU on these devices. Using 30 recent smartphones under controlled acquisition conditions (e.g., no digital zoom), we investigate whether the fingerprint itself is at fault or the way it is compared. We find that a component shared by all fingerprints of the same model inflates the standard signed peak-to-correlation energy (sPCE). To counter this, we propose $Δ$ sPCE: the sPCE between a query image and a candidate camera's fingerprint, minus the mean sPCE between that image and the fingerprints of the remaining candidates, in the spirit of T-norm score normalization. $Δ$ sPCE improves attribution accuracy, suggesting that the device-specific PRNU signal is still present. Although the limited dataset size and the dependence of $Δ$ sPCE on the candidate camera set currently restrict operational applicability, these results suggest that PRNU-based attribution of modern smartphones remains feasible when fingerprints are compared appropriately.

Figures & tables

Smartphone-software-version	Inst.	Ref. img.	Ref. PRNU	Test img.	Match corr.	Mismatch corr.
Google Pixel 10 Pro XL	4	80	4	80	80	240
Google Pixel 10 Pro	6	120	6	120	120	600
Samsung Galaxy S25 Ultra OS SxxR	5	100	5	100	100	400
Samsung Galaxy S25 Ultra OS Sxx5	3	60	3	60	60	120
Samsung Galaxy S25 Ultra OS SxxB	2	40	2	40	40	40
Apple iPhone 17 Pro Max iOS 26.2	4	80	4	80	80	240

Table 1: Description of the dataset and experiment statistics. The dataset consists of a total of 1 200 images acquired with 30 smartphones of 8 different software versions, among three brands. Mismatching correlations of rotated test images are not counted. For example, for the Google Pixel 10 Pro XL, 4 instances and 40 images per smartphone are collected; these images are split into 20 references images and 20 test images per instance, leading to a total of

4\times 20=80

reference images and

4\times 20=80

test images; a reference PRNU is computed per instance, counting 4 in total; for each of the 80 test images, 80 matching correlations can be computed, relative to the corresponding PRNU reference of the same smartphone; for each of the 80 test images,

3\times 80=240

mismatching correlations can be computed relative to the PRNU references of the 3 other smartphones of the same class. For the Samsung software version, SxxR, Sxx5, SxxB correspond to S938BXXS7BYLR, S938USQSBCZF5, S938USQS8BZBB, respectively.

Camera model	Count of metadata profiles
Google Pixel 10 Pro XL	4 (1), 2 (3) and 1 (10)
Google Pixel 10 Pro	6 (1), 2 (1) and 1 (9)
Samsung Galaxy S25 Ultra	5 (1), 3 (1), 2 (9) and 1 (41)
Apple iPhone 17 Pro Max	4 (2), 3 (1), 2 (3) and 1 (42)

Table 2: Count of users per smartphone model metadata profile with at least 40 images, including 20 portrait ones.

X(Y)

means that there are

Y

metadata profiles common to

X

users.

Figure 1: sPCEs computed per smartphone-software-version. sPCEs between a test image noise residual and an estimated PRNU of the same user are reported in green and sPCEs between a test image noise residual and an estimated PRNU of a different user are reported in red . The sPCE threshold per software version inferred from 5% FPR are drawn in blue . While sPCEs between a test image and an estimated PRNU of the same user are a bit higher than sPCEs between a test image and estimated PRNUs of different users, the Apple iPhone 17 Pro Max show very similar distributions. As a result, setting a sPCE threshold to separate the best both distributions is ineffective.

Figure 2: sPCEs computed for Google Pixel 10 Pro smartphones. sPCEs between a test image noise residual and an estimated PRNU of the same user are reported in green ; sPCEs between a test image noise residual and an estimated PRNU of a different user are reported in red . From one instance to another, there is a very similar distribution of sPCEs between test images and the estimated PRNUs of different users. However, the distribution of sPCEs between test images and the same user’s estimated PRNU varies widely from one instance to another, even when accounting for the two sPCE ranges. This difference is especially visible in instance 3.

Figure 3:

\Delta

sPCEs computed per smartphone-software-version.

\Delta

sPCEs testing the correct source camera are denoted in green , while

\Delta

sPCEs testing an incorrect source camera are denoted in red . The

\Delta

sPCE threshold per software version inferred from 5% FPR are drawn in blue . Apple iPhone 17 Pro Max

\Delta

sPCEs testing the correct source camera are significantly higher than

\Delta

sPCEs testing an incorrect source camera. As presented in Table 3 , this substantially improves the separability of both same source and different source distributions thanks to a

\Delta

sPCE threshold compared to a sPCE threshold shown in Fig. 2 .

Figure 4: ROC curves computed from Google, Samsung and Apple smartphones. The true positive rate with a 5% false positive rate is denoted by a dot. Green ROC curves rely on a

\Delta

sPCE threshold and blue ROC curves rely on a sPCE threshold.

	sPCE		$\Delta$ sPCE
Smartphone software version	TPR	$t$	TPR	$t_{\Delta}$	TPR difference
Google Pixel 10 Pro XL	52.5	2 051	77.5	483	25.0
Google Pixel 10 Pro	50.8	2 652	60.0	908	9.2
Samsung Galaxy S25 Ultra OS S938BXXS7BYLR	30.0	1 117	44.0	440	14.0
Samsung Galaxy S25 Ultra OS S938USQSBCZF5	51.7	1 802	63.3	828	11.6
Samsung Galaxy S25 Ultra OS S938USQS8BZBB	47.5	719	70.0	68	22.5

Table 3: Observed TPR (in %) for 5% FPR. When considering a

\Delta

sPCE threshold instead of a sPCE one, all software versions of all models have a significantly higher TPR. For instance, the Apple iPhone 17 Pro Max iOS 26.1 TPR increases from 11.2% to 93.8% by leveraging the

\Delta

sPCE threshold.

\Delta

sPCE TPR - sPCE TPR is denoted as TPR difference measured in %..

Digital Image Forensics

Digital image forensics develops methods to detect, localize, explain, and trace image manipulation, synthetic content, and hidden data.

See all 111 papers

Evaluating Deep Learning and Traditional Approaches Used in Source Camera Identification
Nov 24, 2025Mansur OzamanDigital Image Forensics
Nazarbayev Intellectual School of Science and Mathematics in Medeu District of Almaty
Perceptible or Not? Diagnosing Passive Fingerprints for Speech Deepfake Attribution
Sep 1, 2026Yupei Li, Qiyang Sun, Emmanouil Benetos +2Audio Deepfake Detection Fingerprint Identification
1Imperial College London · 4Technical University of Munich · 2Queen Mary University of London +1
A simple approach for biometrics: Finger-knuckle prints recognition based on a Sobel filter and similarity measures
May 17, 2026E. O. Rodrigues, T. M. Porcino, Aura Conci +1Biometric Identification
*Department of Computer Science, Universidade Federal Fluminense, Rua Edmundo March s/n - Boa Viagem, Niterdi - RJ, 24435-030, Brazil. · Department of Electrical Engineering, Universidade Federal do Maranhão, Av. dos Portugueses s/n - São Luiz - MA, 65085-580, Brazil.

Nov 24, 2025J/K move · Enter open · S save

Evaluating Deep Learning and Traditional Approaches Used in Source Camera Identification

Mansur Ozaman

Nazarbayev Intellectual School of Science and Mathematics in Medeu District of Almaty

Digital Image Forensics

Open paper AskPDF arXiv

Identifying the device that captured a digital image is a central task in image forensics, made more urgent by pervasive smartphone photography and the rise of AI-generated imagery. This paper compares three source camera identification (SCI) approaches - Photo Response Non-Uniformity (PRNU) analysis, JPEG compression artifact analysis, and a convolutional neural network (CNN) - on images from four smartphone devices (Xiaomi Redmi Note 10S, Samsung Galaxy S24, iPhone 13 Pro Max, and iPhone 17 Pro). For each method, discriminative features are extracted (DCT coefficient statistics for JPEG artifacts, denoised noise residuals for PRNU, and learned convolutional features for the CNN) and classified with an SVM (JPEG, PRNU) or trained end-to-end (CNN), using an identical fixed, stratified 70/30 train-test split for direct comparability. The DCT/JPEG-artifact-inspired classifier achieves the highest accuracy (0.90), followed by the PRNU-inspired method (0.71), while the CNN baseline reaches only 0.29. Under this experimental configuration, the DCT/statistical and PRNU-inspired approaches outperform the evaluated CNN baseline. The results motivate evaluation on larger, format-balanced datasets and investigation of hybrid or learned approaches under more robust validation protocols.

Abstract

Digital Image Forensics

Explore similar work

Evaluating Deep Learning and Traditional Approaches Used in Source Camera Identification

Perceptible or Not? Diagnosing Passive Fingerprints for Speech Deepfake Attribution

A simple approach for biometrics: Finger-knuckle prints recognition based on a Sobel filter and similarity measures