An efficient model using deep convolutional neural networks for modeling underwater images

Underwater photography suffers from the dispersion of light in water. Blurring and color distortion of an image are difficult aspects of underwater image analysis. Various methods have been investigated for providing solutions for underwater image restoration. Nevertheless, these approaches still exhibit a regression. Herein, we describe a deep convolutional neural network (DCNetUI) for medium transmission estimation. DCNetUI adopts Deep Convolutional Neural Networks (DCNN), whose layers are specially designed to embody the established assumptions/priors in image restoration. Specifically, layers of Maxout units are used to extract almost all haze-relevant features. Parallel convolution with multi scale features are used to remove haze. Max-polling layer is able to preserve resolution of the feature maps. We also propose a novel nonlinear activation function Bilateral Rectified Linear Unit (BReLU), which alleviate the noisy problem. To refine the transmission map gradient filter is used to smooth the image. Finally, the performances of the proposed and existing methods were verified by comparing the experimental results with those of known methods under quality metric settings. Experiments on benchmark images show that DCNetUI achieves superior performance over existing methods, yet keeps efficient and easy to use.The recommended method enhances the color of an image by removing the influence of the aquatic elements. It increased the SSIM by 29%, with a value of 0.967 and PSNR of 47%, with a maximum value of 54.537.