A Multi-Domain Feature Fusion Framework for Generalizable Deepfake Detection Across Different Generators

Authors: Amna Amjid, Sana Qadir, Mehwish Fatima, Raja Khurram Shahzad

Published: 2026-06-12 08:12:03+00:00

AI Summary

This paper proposes SGFF-Net (Spatial-Gradient-Frequency Fusion Network), a multi-domain deepfake detection framework that integrates spatial, gradient, and DWT-based frequency representations within a dual residual learning architecture. It addresses the challenge of generalizing deepfake detection across different generative models, including GANs and diffusion models, and systematically evaluates cross-generator robustness. The framework significantly improves detection performance and generalization, especially when combined with multi-source training and data augmentation.

Abstract

Deepfakes are artificially generated images, audio, or videos that threaten privacy, security, and information integrity. Detecting such content is crucial for countering disinformation, as the latest models generate highly realistic content. While spatial- or frequency-based approaches achieve good detection rates on Generative Adversarial Networks (GANs)-based generated deepfakes, they often struggle with recent diffusion model-generated images. In particular, existing approaches rarely exploit complementary multi-domain representations or systematically evaluate cross-generator robustness. To address these challenges, we propose a multi-domain deepfake detection framework called SGFF-Net (Spatial-Gradient-Frequency Fusion Network) that integrates spatial, gradient, and DWT (Discrete Wavelet Transform)-based frequency representations within a dual residual learning architecture. Experimental results show that the SGFF-Net achieves 98.95\\% accuracy in intra-dataset evaluation and improves performance in both cross-model (70.46\\%) and cross-paradigm (69.94\\%) settings. Incorporating multi-source training and data augmentation further enhances robustness, increasing accuracy from 70.46\\% to 79.80\\% in cross-model evaluation, from 69\\% to 78\\% in cross-paradigm evaluation, and from 61.50\\% to 75.80\\% on real-world data. Unlike single-domain detectors, the SGFF-Net learns complementary forensic cues across spatial, gradient, and wavelet-frequency domains, resulting in greater robustness under cross-generator and cross-paradigm evaluation. The results further show that combining multi-domain representations with data diversity and augmentation substantially improves generalization, providing practical insights for developing more reliable deepfake detection systems.

Key findings

SGFF-Net achieved 98.95% accuracy in intra-dataset evaluation and significantly improved performance in cross-model (70.46%) and cross-paradigm (69.94%) settings. Incorporating multi-source training and data augmentation further boosted robustness, increasing cross-model accuracy to 79.80% and cross-paradigm accuracy to 78%, and reaching 75.80% accuracy on real-world data, demonstrating superior generalization over single-domain detectors.

Approach

The SGFF-Net framework extracts spatial (RGB), gradient, and DWT-based frequency representations in parallel. These multi-domain features are then fused and processed by a Dual Residual Network (DRN) for classification. Data augmentation and multi-source training strategies are employed to enhance generalization across various deepfake generators and paradigms.

Datasets

DeepFakeFace (DFF) (diffusion-based), Diffusion Face (DiffFace) (diffusion-based), Diverse Fake Face Dataset (DFFD) (GAN-based), MIX_ALL (combined DFF_C, DiffFace_A, DFFD), Real_World (internet-sourced unseen images). Source datasets for real images include IMDb-Wiki Dataset, CelebA, and FFHQ.

Model(s)

SGFF-Net (Spatial-Gradient-Frequency Fusion Network), Dual Residual Network (DRN) for detection, ResNet-50-based transformation model for gradient maps, Multi-task Cascaded Convolutional Networks (MTCNN) for face extraction and localization.

Author countries

Pakistan, Sweden

← Previous