Frequency-Spatial Feature Fusion Network for Infrared and Visible Image Fusion

Abstract

Infrared and visible image fusion seeks to retain complementary information from source images and generate a comprehensive image. Most fusion methods ignore the detailed information in the frequency domain. To address this problem, we propose a Frequency-spatial Feature Fusion Network (F3Net) in this work. The F3Net consists of three modules, namely Frequency-Spatial Feature Extraction Module (FSFEM), Feature Fusion Module (FFM), and Image Reconstruction Module (IRM). First, the FSFEM is built to extract complementary information from the source image separately in the frequency and spatial domains. Then, the FMM is introduced to fuse the features of the frequency and spatial domains. Finally, the fused image is reconstructed by IRM. Comprehensive experiments demonstrate that the F3Net outperforms the state-of-the-art (SOTA) methods subjectively and objectively.