Mdbfusion: A Visible And Infrared Image Fusion Framework Capable For Motion Deblurring

As a vital technology in the field of image processing, numerous visible and infrared fusion methods have emerged with the aim of generating a single image containing salient targets and abundant details. However, owing to the influence of target motion or camera shake, source images may suffer from motion blur, resulting in blurred edges and unrecognizable objects. To address this issue, this paper bridges the gap between deblurring and fusion tasks and proposes a joint motion deblurring and fusion network for visible and infrared image (MDbFusion). On the one hand, we innovatively merge motion deblurring task into network design, which effectively ensures the capacity of MDbFusion to process images in the extreme condition. On the other hand, an Adaptive Weight Module (AWM) is designed to calculate contribution between visible and infrared features, which solves the channel contrast between two tasks, greatly reducing the complexity of network. Extensive experiments demonstrate that the MDb-Fusion outperforms state-of-the-art (SOTA) fusion algorithms in terms of preserving texture details and quantitative metrics. Furthermore, we also compare with a two-step fusion strategy that first deblurring then fusion, both with SOTA methods. The results reveal superiority of our framework in coupling and reciprocity between two tasks.