Gaussian Image Denoising Method Based on the Dual Channel Deep Neural Network with the Skip Connection

Abstract

In the era of rapid development of artificial intelligence technology, image denoising methods based on deep learning have achieved better and better results, and some deeper networks have also been proposed. However, with the increasing number of network layers, gradient explosion and over fitting problems also appear in the training process. In this paper, a new Gaussian image denoising method based on dual channel deep neural network with skip connection is proposed. The network is composed by the first layer network and the second layer network in parallel, so as to widen the width of the network. It not only improves the denoising effect, but also reduces the problems in the training process. The first layer uses dilated convolution to expand the receptive field of the network, and the second layer is composed of skip connection modules. The method is tested on the data set68 and achieves good results.