Enhanced Deep Learning DeepFake Detection Integrating Handcrafted Features

View on arXiv ← Back to list

Authors: Alejandro Hinke-Navarro, Mario Nieto-Hidalgo, Juan M. Espin, Juan E. Tapia

Published: 2025-07-28 08:19:22+00:00

AI Summary

This research proposes a deepfake detection framework that integrates handcrafted frequency-domain features (DCT, SRM, DFT, ELA, SVD) with RGB inputs. This hybrid approach aims to improve generalization across different datasets by leveraging both spatial and frequency-domain artifacts introduced during image manipulation.

Abstract

The rapid advancement of deepfake and face swap technologies has raised significant concerns in digital security, particularly in identity verification and onboarding processes. Conventional detection methods often struggle to generalize against sophisticated facial manipulations. This study proposes an enhanced deep-learning detection framework that combines handcrafted frequency-domain features with conventional RGB inputs. This hybrid approach exploits frequency and spatial domain artifacts introduced during image manipulation, providing richer and more discriminative information to the classifier. Several frequency handcrafted features were evaluated, including the Steganalysis Rich Model, Discrete Cosine Transform, Error Level Analysis, Singular Value Decomposition, and Discrete Fourier Transform

Key findings

Minimum score-level fusion of RGB and DCT features yielded the best performance. Integrating handcrafted frequency features significantly improved deepfake detection accuracy and generalization across diverse datasets compared to using only RGB inputs. The DCT feature was identified as particularly effective.

Approach

The approach combines handcrafted frequency-domain features (e.g., DCT, SRM) with conventional RGB image inputs. These features are then fed into EfficientNetV2 B0 or MobileViT-S models, and a minimum score-level fusion is applied to combine the outputs of different models for improved detection accuracy.

Datasets

FaceForensics++, Celeb-DF, DeepfakeTIMIT, DeePhy, Defacto, SWAN-DF

Model(s)

EfficientNetV2 B0, MobileViT-S

Author countries

Spain, Germany

← Previous