Work

ULightIF: Unsupervised Low Light Infrared and Visible Image Fusion Method

Year: 2026

Type: article

Abstract: Current advanced research on infrared and visible image fusion (IVIF) primarily focuses on improving fusion performance under normal illumination, while its applicability in low-light conditions remains underexplored. To address this, we propose ULightIF, an unsupervised IVIF framework tailored for low-light scenarios. Our method innovatively leverages the pre-enhanced results from the “enhance first, fuse later” process to guide the “fuse first, enhance later” procedure, thereby overcoming the limitations of existing low-light fusion methods that typically rely on a single strategy. From a structural perspective, ULightIF integrates low-light enhancement, multimodal image fusion, and image decomposition subnetworks within a flexible architecture. It further adopts a convolutional re-parameterization technique, resulting in a lightweight network with only approximately 0.0597 M parameters. The network is trained using a carefully designed set of unsupervised loss functions, using only paired low-light infrared and visible images without requiring normally illuminated data. Extensive experiments demonstrate that our method achieves a favorable balance between high-quality visual performance and adaptability to downstream tasks, while maintaining high efficiency for practical deployment.

Source: IEEE Transactions on Multimedia

DOI: 10.1109/tmm.2026.3723605

Authors: Muhang Cheng, Haiyan Huang, Hongwei Mo

Institutions: Harbin Engineering University, Wuhan University

Language: English

Topic: Advanced Image Fusion Techniques

Subfield: Media Technology

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: hybrid

Estimated APC paid: $2,645