LGIFNet: an infrared and visible image fusion network with local-global frequency interaction

The fusion of infrared and visible light images aims to combine the texture details of visible light and the thermal radiation information of infrared images to generate a more comprehensive fused image, providing stronger support for visual analysis and processing. However, existing methods struggle to preserve the integrity of the global structure and the clarity of details, making it difficult to process global features and detail information simultaneously, resulting in loss or blurring of fused image information. To address these challenges, this paper proposes a dual-branch feature extraction and weighted frequency fusion network using Transformer and convolutional neural networks. The central encoder enhances global feature representation, preserving structural integrity and preventing detail loss. A low-frequency LIA-Transformer module combines local importance attention with MLP Transformer to retain global structural features. A high-frequency reversible convolution module ensures detail retention via an inverted residual block. Finally, a weighted frequency fusion module adaptively fuses low- and high-frequency features through dynamic weight allocation to enhance detail clarity and information integrity. Experiments on multiple datasets show the proposed method outperforms existing ones in brightness preservation, edge clarity, and contrast, effectively improving fused image quality and reliability.