Title: FARL: deepfake detection based on adaptive feature aggregation and reconstruction learning

Authors: Min Long; Chunbin Chen; Le-Bing Zhang; Fei Peng

Addresses: School of Electronics and Communication Engineering, Guangzhou University, Guangzhou, 510006, China ' School of Computer and Communication Engineering, Changsha University of Science and Technology, Changsha, Hunan, 410114, China ' School of Computer and Artificial Intelligence, Huaihua University, Huaihua, Hunan, 418000, China ' School of Artificial Intelligence, Guangzhou University, Guangzhou, Guangdong, 510006, China

Abstract: Existing face deepfake detection methods mainly rely on extracting specific forgery patterns (such as facial features, noise characteristics, and frequency domain information) to enhance the performance of detection models. However, these learning strategies, which depend on specific forgery patterns, limit the model's generalisation ability when facing diverse forgery types, making it difficult to effectively identify fake images with unknown forgery patterns. To address this issue, a novel deepfake detection framework FARL is proposed based on adaptive feature aggregation and reconstruction learning, which enhances the model's sensitivity to forgery traces, thereby significantly improving detection performance. Specifically, a cross feature aggregation encoder (CFAE) is introduced to enhance the model's sensitivity to forgery traces through an attention refinement block and a cross fusion strategy. Additionally, a block feature aggregation decoder (BFAD) is implemented to aggregate high-resolution feature maps from encoder blocks and low-resolution feature maps from the previous decoder block. Finally, the differences between the original image and the reconstructed image are utilised as a guide. Experimental results on several benchmark datasets demonstrate the effectiveness of the proposed method for deepfake detection.

Keywords: deepfake detection; deep learning; multi-scale; reconstruction learning.

DOI: 10.1504/IJAACS.2026.152866

International Journal of Autonomous and Adaptive Communications Systems, 2026 Vol.19 No.2, pp.275 - 294

Received: 10 Jan 2025
Accepted: 18 Mar 2025
Published online: 13 Apr 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article