Deep retinex decomposition network for underwater image enhancement

This paper introduces a deep retinex decomposition network for underwater image enhancement to conquer the color imbalance, blurring, low contrast, etc. Specifically, we first designed a novel convolutional neural network to estimate the illumination and get reflectance. Then we changed the general idea of processing low light enhancement based on retinex, we perform color balance and illumination correction on the decomposed reflectance and illumination respectively. Finally, the fused reflectance image and illumination image are produced by post processing to get over blurring, etc. The experiments confirm that the proposed method can retain more details and edge information. Meanwhile, compared with other underwater image enhancement methods, the proposed method performs better in terms of visual effects with nearly 20% improvement in objective evaluation.