Learning-Based Dark and Blurred Underwater Image Restoration

Figure 3

The pipelines are based on traditional methods and deep learning. The input test images are taken in dark and blurred water. The goal of the algorithm is to generate the respective high-quality clear images. The top subimage surrounded by the dotted box represents traditional methods, including the pipeline only using an underwater image enhance/restoration algorithm (line A) and the cascade algorithms (line B and line C). The bottom subimage represents an end-to-end algorithm based on deep learning including the training phase (line D_1) and the test phase (line D_2).