Convolutional Neural Network Denoising Method Based on Multisize Features
Xiaoyu Li, Qingwei Gao, Yixiang Lu, Dong Sun · 2018
In order to better remove the noise in the image, this paper uses the neural network to fit the noisy image to the denoised image based on the deep learning principle. For the general neural network denoising problem due to con-volutional features constrained and pooled information lost, the ClonvBlock unit was used to broaden the network width and depth, thereby obtaining multiscale features while reducing the number of training parameters, and then improving the neural network. Trained to output denoised images using deconvolution networks and jump connections. Compared with previous methods of denoising based on deep learning, this method recovers more details of the image in less time-consuming situations and achieves better denoising results.