Dehazing Underwater Images Using Encoder Decoder Based Generic Model-Agnostic Convolutional Neural Network
Rosemol Thomas, Lidiya Lilly Thampi, Suraj Kamal, Arun A. Balakrishnan, T.P. Mithun Haridas, M. H. Supriya · 2021
Enhancing degraded underwater images is a challenging problem for researchers. Deep learning approaches can significantly provide improvement in image restoration and enhancement. Moreover automatic analysis of marine digital data has a high research demand. This paper aims to introduce a fully connected convolutional neural network for dehazing underwater images. The low level and high level features are integrated by an encoder-decoder deep framework which helps to recover the hazy image. The proposed method is evaluated quantitatively and qualitatively on Underwater Image Enhancement Benchmark dataset (UIEB). The model outperforms existing methods in terms of metrics: SSIM, PSNR and MSE and has the ability to dehaze underwater images retaining fine details.