.png){kind=icon}
Return
DIFF-SB: A process-optimized diffusion model for infrared-visible image fusion using Split Bregman
.png){kind=icon}
PRE
AI
DOI：10.1016/j.infrared.2026.106797.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Infrared-Visible image fusion (IVIF) synthesizes complementary thermal and visual information to produce high-contrast images with salient targets and fine textures. Existing diffusion-based IVIF methods suffer from two critical limitations: over-reliance on high-quality fusion priors, and inadequate source information utilization that causes severe information loss in low-light or thermal-dominated scenarios. To address these issues, we propose DIFF-SB, a diffusion fusion framework which incorporates a deep unfolding Split Bregman algorithm and adapts its mathematical modeling for image fusion. It embeds process optimization and explicit physically motivated constraints into the diffusion sampling process, harnessing each modality's inherent properties. DIFF-SB enforces a sparsity-fidelity regulation mechanism through Split Bregman's problem decomposition and iterative optimization. Additionally, we introduce an Attention Feature Decoupling Module to decompose source images into complementary base structure and detail subspaces. This decomposition establishes distinct feature foundations for subsequent constraint enforcement. It effectively overcomes the limitation of conventional Split Bregman optimization, which is confined to a single feature dimension. Consequently, our framework can impose multidimensional, differentiated constraints on cross-modal information. Complementary details are thus preserved with greater precision in complex scenarios. This leads to superior fusion performance and improved interpretability. Extensive experiments on infrared-visible datasets demonstrate that DIFF-SB achieves competitive performance, consequently enhancing downstream tasks such as object detection and semantic segmentation.
Keywords:
Image fusion
Infrared image
Diffusion model
Split Bregman
Journal
I
IF:
3.4
Papers:
530
Citations:
0
Organization
Cited Papers
No cited papers available