Multi scale perceptual underwater image enhancement based on feature fusion network

Due to the absorption and scattering of light, underwater imaging is often degraded by low contrast or color projection. This fact limits the accuracy of object identification in underwater and Marine environments. In this paper, to address this problem, we propose a multi-scale global-local feature fusion network (MGDFN) that progressively improves degraded underwater images in a coarse-to-fine manner. Specifically, our MGDFN network consists of two cascaded subnetworks. The first subnet (an optimized encoder-decoder subnet, OEDNet) learns multi-scale context information through a feature attention module and multi-scale fusion mechanism, while the second subnet (a pyramid-based feature refinement subnet, PFRNet) preserves fine spatial details using a detail enhancement module and a self-supervised recovery module. Deploy tail enhancement modules and supervised recovery modules between subnetworks to further improve image color and detail recovery. Our proposed multi-scale global-local feature Fusion network (MGDFN) is extensively evaluated on popular datasets and achieves state-of-the-art performance. Our method is also more friendly to object recognition than its peers underwater objects are easier to identify when processed by our method.