Authors: Subhrojyoti Mukherjee, Manoranjan Mohanty
Published: 2025-08-27 09:20:56+00:00
AI Summary
This paper proposes a two-factor authentication method for selfie banking to mitigate deepfake attacks. It leverages Photo Response Non-Uniformity (PRNU) analysis to verify the source camera, in addition to facial recognition, thus preventing authentication using deepfakes generated from stolen images.
Abstract
Fake images in selfie banking are increasingly becoming a threat. Previously, it was just Photoshop, but now deep learning technologies enable us to create highly realistic fake identities, which fraudsters exploit to bypass biometric systems such as facial recognition in online banking. This paper explores the use of an already established forensic recognition system, previously used for picture camera localization, in deepfake detection.
Key findings
Deepfake videos successfully bypassed liveness detection but failed the PRNU camera authentication test. The proposed two-factor authentication scheme effectively prevents deepfake-based authentication in selfie banking scenarios. The PRNU method proved robust against deepfakes generated from different cameras.
Approach
The approach uses PRNU-based camera authentication as a second factor alongside facial recognition. A camera fingerprint is created during registration and compared to a fingerprint from the authentication attempt. A match indicates the image comes from a registered camera, preventing deepfake authentication.
Datasets
UNKNOWN. The paper mentions using images and videos from a laptop camera, smartphone camera, and webcam, as well as deepfakes generated using the Akool software. No publicly available dataset is explicitly named.
Model(s)
Akool (for deepfake generation), an in-house Python code (for liveness detection), and PRNU analysis (for camera identification). No specific deep learning model architecture for deepfake detection is mentioned besides the components within Akool.
Author countries
India, Qatar