SIEFusion: Infrared and Visible Image Fusion via Semantic Information Enhancement

At present, most of existing fusion methods focus on the metrics and visual effects of image fusion, but ignore the requirements of high-level tasks after image fusion, which leads to the unsatisfactory performance of some methods in subsequent high-level tasks such as semantic segmentation and object detection. In order to address this problem and obtain images with rich semantic information for subsequent semantic segmentation tasks, we propose a fusion network for infrared and visible images based on semantic information enhancement named SIEFusion. In our fusion network, we design a cross-modal information sharing module(CISM) and a fine-grained detail feature extraction module(FFEM) to obtain better fused images with more semantic information. Extensive experiments show that our method outperforms the state-of-the-art methods both in qualitative and quantitative comparison, as well as in the subsequent segmentation tasks.