Multisensor Fusion of Infrared and Visible Images via Latent Low-Rank Representation and Saliency-Based Decomposition
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
In the application of multi-sensor array cameras, the fusion of infrared and visible images can enhance the multi-modal perception of system and compensate for the lack of reliability caused by single spectrum information. Nowadays, infrared and visible image fusion is a key technology in systems such as aerospace, autonomous, and surveillance systems. Traditional methods including sparse and low-rank representations (LRR) provide interpretability but often lack robustness when facing complex noise and structural variations. Although deep learning-based methods have made advancements, these methods still encounter difficulties in aligning cross-modal features and achieving robust generalization. Motivated by the limitations, this paper proposes an innovative fusion method that integrates enhanced latent low-rank representation (LatLRR) with saliency-based decomposition, called LRR-SD. The framework introduces an L2;1-norm constraint to improve robustness of LatLRR. Besides, the proposed method adopts the alternating direction method of multipliers (ADMM) for optimization algorithm. Then, a three-layer decomposition architecture is designed to separate saliency, detail, and base components, thereby adaptively preserving thermal targets and texture information. Results from experiments on several available datasets indicate that LRR-SD effectively retains infrared targets while maintaining natural visible background details, demonstrating advantages over existing techniques in both qualitative and quantitative evaluations.
Fields of Study & Subjects
Advanced Image Fusion Techniques
Image Enhancement Techniques
Advanced Vision and Imaging
Publication details
- Journal / source
- IEEE Sensors Journal
- Volume
- 26
- Issue
- 9
- Pages
- 13573–13588
- Publication date
- 26 March 2026
- Language
- EN