Model-Based Underwater Image Simulation and Learning-Based Underwater Image Enhancement Method

Due to the absorption and scattering effects of light in water bodies non-uniformity insufficiency artificial illumination, underwater images often present various degradation problems, impacting their utility applications. In this paper, we propose a model-based image simulation learning-based enhancement method for coping with problems images. We first derive simplified model describing images, then that can generate wide range parameter values. The proposed also comes an image-selection part, which helps prune dataset so it serve as training set learning enhance targeted Afterwards, convolutional neural network based on encoder-decoder backbone learn from simulated Experiments real different demonstrate effectiveness method, reveal advantages comparison many state-of-the-art methods.