A Progressive Frequency-Guided Decomposition Method for Infrared and Visible Image Fusion

Abstract

Infrared and visible image fusion seeks to combine complementary information acquired by infrared and visible sensors, simultaneously highlighting salient targets and preserving texture details in a fused image. Existing decomposition-based fusion methods can separate shared and modality-specific features, but they insufficiently organize and leverage cross-modal complementary information progressively and coherently, resulting in unstable structural preservation, blurred texture details, and weakened target prominence. To address these issues, this article proposes a progressive frequency-guided decomposition fusion network (PFGD-Net). The method first establishes a dual-branch feature decomposition framework using a structure–detail constraint (SDC) that enforces structural consistency and decouples detail features, enabling an initial separation of cross-modal representations. A cross-modal frequency selection (CFS) and re-decomposition mechanism is then introduced to dynamically select and reorganize complementary information. A frequency-conditioned adaptation (FCA) module further applies differential modulation to low-frequency structural features and high-frequency detail features. Additionally, heterogeneous enhancement is applied according to the characteristics of each branch, with a structural global–local interaction (SGI) block in the shared branch and a detail high-order interaction (DHI) block in the detail branch, improving structural preservation, detail restoration, and target saliency. Experiments on three public datasets, including TNO, RoadScene, and MSRS, demonstrate that PFGD-Net achieves consistently superior performance on four commonly used evaluation metrics and produces visually favorable results with a better balance among target saliency, texture preservation, and structural consistency. These results verify the effectiveness of the proposed method for decomposition-based infrared and visible image fusion.