An Infrared and Visible Image Fusion Network Based on Frequency Domain Information Extraction

The objective of infrared and visible light image fusion is to integrate data from distinct sensors to generate a more comprehensive and accurate representation of a scene. In this study, we introduce FDFusionNet, a fusion framework based on frequency domain information extraction. FDFusionNet effectively preserves thermal targets from infrared images, as well as texture structures and background details from visible images. The framework segments the extracted image features into spatial domain feature extraction and frequency domain feature extraction for enhanced performance. Fourier transform facilitates the extraction of amplitude and phase components in the frequency domain, integrating background information, while spatial domain features are extracted using SEDense. This paper achieves a balance between global and local information through the complementary nature of frequency and spatial domain features. Moreover, we incorporate information entropy to design different modal weights, effectively controlling the input ratios and suppressing redundant information and noise. Extensive qualitative and quantitative experiments demonstrate that our fusion algorithm effectively balances texture and structural information, outperforming other fusion methods on the TNO public dataset. Our method consistently ranks among the top four across various evaluation metrics on TNO data, validating its superior performance in infrared and visible light fusion scenarios.