An Underwater Image Enhancement Fusion Network

To deal with the challenging underwater vision problem, we propose an underwater image enhancement fusion network, which uses a data-driven neural network to implement a fusion principle. Specifically, this network utilizes residual learning and multi-scale fusion modules to complete the fusion process. The numerical experiments over different underwater image datasets demonstrate that the proposed network consistently produces satisfied enhancement results compared to state-of-the-arts including both model-based and data-driven counterparts. Furthermore, the proposed network significantly reduces the neuron consumption that improves its practical applicability especially for the resource-limited underwater environment.