Underwater image enhancement with encoding-decoding deep CNN networks

Image enhancement is an important topic in the field of machine vision and image processing. Complex underwater environment poses great challenge for machine vision due to the turbid water medium. Serious scattering and absorption make the underwater images have complicated noise distribution. This paper proposes an underwater image enhancement model based on Encoding-Decoding deep CNN networks. We employ the convolution layers as encoding while deconvolution layers as decoding. The model achieves the image enhancement in an end-to-end adaptive way rather than considering the physical environment. We provide several comparison experiments with different datasets. Experiments show that it outperforms state-of-the art underwater image restoration methods in underwater image defogging, denoising and color enhancement. Finally, the model is employed to handle the underwater images with the different levels of noise and shows good performance.