Underwater Image Deblurring Framework Using A Generative Adversarial Network

Blur effects commonly exist in the underwater images due to the camera shake, object motion and insufficient light in the practical underwater environment. They greatly affect the accuracy and efficiency of underwater investigations (an AUV-based vision exploration). To address this issue, we propose a solution using deep learning-based framework to translate blurry underwater images into clear ones. The network consists of two sub-networks, they incorporate to solve the notorious image translation problem by learning the blur kernel priors and the color priors. A specific large-scale underwater deblurring dataset is selected for training and evaluating our deep framework. Extensive experiments show our framework exhibits satisfying results.