Boosting target-level infrared and visible image fusion with regional information coordination

Abstract

The target-level infrared and visible image fusion aims to prominently retain the feature of target areas in the entire scene of fusion results. Nonetheless, most of existing fusion methods tend to evaluate the global information and ignore the retention of specific target information during feature extraction. A few existing target-level fusion methods also have the problem of missing target or scene information under special conditions. In order to address the challenges of image fusion and make further deployment planning in high-level image vision tasks, we propose a target-level infrared and visible image fusion method. In our method, a scene texture attention module is designed to enhance the complementary description of global scene information, and a target extraction module with the target-level loss function is designed to prominently retain the feature of target areas. Furthermore, target information and scene information are equilibrated by the coordination of target-scene information loss function. A large number of comparison experiments with the state-of-the-art methods demonstrate that our fusion method has competitive advantages in highlighting target features and describing global scene. More importantly, in downstream target detection and depth estimation tasks, the excellent performance in accuracy and speed considerably enhances the portability of our method.