ASMF: A Self-Supervised Atmospheric Scatter Model-Based Fusion Network for Infrared Image Enhancement

Infrared camera sensors typically capture thermal radiation data, which can be influenced by random scattering and reflection during transmission. As a result, the infrared images produced often exhibit low contrast and clarity. This problem is especially evident in uncooled long-wave infrared (LWIR) images, which experience a decline in image quality, making it more difficult to enhance contrast. To tackle this issue, we propose a self-supervised network model that utilizes an atmospheric scattering model (ASM) specifically designed for the distinct characteristics of infrared imaging. Moreover, leveraging the analysis of ASM and the characteristics of infrared imaging, we introduce an image enhancement framework that integrates multi-scale dehazing with fusion techniques to facilitate self-supervised learning. Our method includes two self-supervised sub-networks, ASM-Net and F-Net, each equipped with tailored loss functions to address the challenges arising from the limited availability of high-quality datasets and the variability in imaging from different manufacturers. We improve self-supervised learning by integrating degradation models into the network architecture and incorporating prior knowledge into the loss functions. To evaluate our proposed method, we have developed a new LWIR dataset featuring real-world scenes. Extensive experiments conducted on established benchmark datasets, as well as our dataset, demonstrate that our approach significantly surpasses existing methods in both qualitative and quantitative assessments. The source code and dataset are available at https://github.com/ANDYLLY/Infrared-image-enhancement-method-ASMF-and-LWIR-Dataset-LGC.