Improved Parallel Residual Dense Denoising Network Based on Multi-scale Fusion
WANG Jie1,LUO Jing-rui1,YUE Guang-de2
1(School of Automation and Information Engineering,Xi'an University of Technology,Xi'an 710048,China)2(School of Mathematics and Statistics,Xi'an Jiaotong University,Xi'an 710049,China)
Abstract：Convolutional neural network has achieved good results in image denoising.However,the traditional compression-decompression neural network inevitably damages the information in the original image.In order to more effectively remove the noise from the image,this paper proposed an improved multi-scale feature fusion based parallel residual dense denoising network framework to better recover the edge and texture information of the image.First,the parallel structure was used to combine the image features with different depths.Each branch of the network consists of some residual dense blocks.On this basis,the remote skipping connections between the RDB were added to overcome the gradient disappearance and gradient dispersion problems during the network training and improve the performance of the network.In addition,on the basis of combining the shallow and deep information of the image,the multi-scale feature fusion blocks were added in each branch of the network to obtain multi-scale image features with different depths.Finally,the performance of the network was further improved by using residual learning.Comparative experiments show that the proposed network has achieved good results under different noise intensity,which further proved that the proposed network can effectively retain the edge and texture information of the original image while noise suppression.
2021, Vol. 42 
