Robust Deepfake Detection for Electronic Know Your Customer Systems Using Registered Images

View on arXiv ← Back to list

Authors: Takuma Amada, Kazuya Kakizaki, Taiki Miyagawa, Akinori F. Ebihara, Kaede Shiohara, Toshihiko Yamasaki

Published: 2025-07-30 12:16:27+00:00

AI Summary

This paper introduces a deepfake detection algorithm for eKYC systems that leverages temporal inconsistencies in identity vectors from a face recognition model and compares these vectors to those from a registered image. This approach improves detection accuracy and robustness against image degradation.

Abstract

In this paper, we present a deepfake detection algorithm specifically designed for electronic Know Your Customer (eKYC) systems. To ensure the reliability of eKYC systems against deepfake attacks, it is essential to develop a robust deepfake detector capable of identifying both face swapping and face reenactment, while also being robust to image degradation. We address these challenges through three key contributions: (1)~Our approach evaluates the video's authenticity by detecting temporal inconsistencies in identity vectors extracted by face recognition models, leading to comprehensive detection of both face swapping and face reenactment. (2)~In addition to processing video input, the algorithm utilizes a registered image (assumed to be genuine) to calculate identity discrepancies between the input video and the registered image, significantly improving detection accuracy. (3)~We find that employing a face feature extractor trained on a larger dataset enhances both detection performance and robustness against image degradation. Our experimental results show that our proposed method accurately detects both face swapping and face reenactment comprehensively and is robust against various forms of unseen image degradation. Our source code is publicly available https://github.com/TaikiMiyagawa/DeepfakeDetection4eKYC.

Key findings

The proposed method outperforms baselines in both in-dataset and cross-dataset evaluations, particularly demonstrating improved robustness against various image degradations. Using a higher-performance face recognition model significantly enhances both detection performance and robustness to image degradation.

Approach

The algorithm compares identity vectors extracted from input video frames with those from a registered image (assumed genuine) to identify inconsistencies. It also analyzes temporal inconsistencies within the video's identity vectors using a recurrent neural network for classification. A robust face feature extractor enhances performance and robustness to image degradation.

Datasets

Korean DeepFake Detection Dataset (KoDF), Celeb-DF v2, DeepFakeDetection (DFD), DeepFake Detection Challenge Preview (DFDCp), FaceForensics++ (FF++) (mentioned but not used for training/evaluation in main contribution)

Model(s)

Bidirectional Gated Recurrent Unit (GRU) with ResNet100 (pre-trained with AdaFace loss on WebFace12M, MS1MV2; also with ArcFace loss on MS1MV2 for comparative experiments), EfficientNet-B4 (mentioned in related work as used by baseline methods)

Author countries

Japan

← Previous