Image enhancement plays an important role in the field of underwater vision. Numerous underwater image enhancement algorithms have been proposed in the last few years, which have achieved some good results in processing specific underwater images. However, the effectiveness of these algorithms to cope with different underwater environments remains uncertain. To address this issue, we propose a water body classification label based on scattering characteristics and construct a dataset with a large number of photos of experiments in different water conditions. Meanwhile, based on different types of water bodies we also trained a network model which has thirteen classifications. Using this dataset, we study comprehensively these underwater image enhancement algorithms qualitatively and quantitatively and match each type of underwater image with an optimal underwater image enhancement algorithm. An underwater image enhancement algorithm based on deep-learning water pre-classification is then proposed. This adapted algorithm is applied to process real underwater images captured by the underwater robot and obtains good processing results. It also contributes to further research on underwater image enhancement.
|