Article: Attentional dual-branch shallow feature enhancement and gated fusion for improved image copy-move forgery detection Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.55 pp.1 - 23 Abstract: Detecting subtle tampering traces within complex backgrounds remains a significant challenge in image copy-move forgery detection, primarily due to the inadequacy of shallow feature extraction. To overcome these limitations, this paper proposes an enhanced DeepLabV3+ architecture designed for efficient multi-scale feature fusion. The framework utilises a lightweight MobileNetV3 backbone within an encoder-decoder structure, integrated with an improved atrous spatial pyramid pooling (ASPP) module employing depthwise separable dilated convolutions. To strictly preserve low-level details, we introduce a dual-branch shallow feature enhancement module (dual-branch SFEM) augmented by efficient channel attention (ECA). Furthermore, the feature fusion stage is optimised through architectural restructuring to reduce computational complexity while maintaining performance. A key innovation is the inclusion of a lightweight gating network that generates spatially adaptive weights, dynamically balancing the trade-off between semantic abstraction and detail preservation. Extensive experiments on the CASIA, DEFACTO, and COVERAGE datasets demonstrate the model's superiority over state-of-the-art methods. Specifically, the proposed method achieves an AUC of 95.41% and an F1 score of 77.24% on the DEFACTO dataset, while exhibiting robust generalisation capabilities on CASIA 1.0 (AUC: 78.93%, F1: 57.68%). Inderscience Publishers - linking academia, business and industry through research

Title: Attentional dual-branch shallow feature enhancement and gated fusion for improved image copy-move forgery detection

Authors: Zirui Qi; Yilihamu Yaermaimaiti; Fusheng Zhao

Addresses: School of Intelligent Science and Technology, Xinjiang University Urumqi, Xinjiang, 830017, China ' School of Electrical Engineering, Xinjiang University Urumqi, Xinjiang, 830017, China ' School of Intelligent Science and Technology, Xinjiang University Urumqi, Xinjiang, 830017, China

Abstract: Detecting subtle tampering traces within complex backgrounds remains a significant challenge in image copy-move forgery detection, primarily due to the inadequacy of shallow feature extraction. To overcome these limitations, this paper proposes an enhanced DeepLabV3+ architecture designed for efficient multi-scale feature fusion. The framework utilises a lightweight MobileNetV3 backbone within an encoder-decoder structure, integrated with an improved atrous spatial pyramid pooling (ASPP) module employing depthwise separable dilated convolutions. To strictly preserve low-level details, we introduce a dual-branch shallow feature enhancement module (dual-branch SFEM) augmented by efficient channel attention (ECA). Furthermore, the feature fusion stage is optimised through architectural restructuring to reduce computational complexity while maintaining performance. A key innovation is the inclusion of a lightweight gating network that generates spatially adaptive weights, dynamically balancing the trade-off between semantic abstraction and detail preservation. Extensive experiments on the CASIA, DEFACTO, and COVERAGE datasets demonstrate the model's superiority over state-of-the-art methods. Specifically, the proposed method achieves an AUC of 95.41% and an F1 score of 77.24% on the DEFACTO dataset, while exhibiting robust generalisation capabilities on CASIA 1.0 (AUC: 78.93%, F1: 57.68%).

Keywords: image forgery detection; gated fusion; efficient channel attention mechanism; dual-branch shallow feature enhancement module; DB-SFEM.

DOI: 10.1504/IJICT.2026.153798

International Journal of Information and Communication Technology, 2026 Vol.27 No.55, pp.1 - 23

Received: 19 Jan 2026
Accepted: 23 Mar 2026
Published online: 26 May 2026 *

Title: Attentional dual-branch shallow feature enhancement and gated fusion for improved image copy-move forgery detection

Keep up-to-date