Residual Learning of Feedforward Denoising ResNet for Image Denoising
Chuanhui Shan, Hu Li · 2024
As deep learning has advanced, it has been used extensively to the subject of image denoising and has proven to be a successful method. Based on the excellent performance of ResNet, this paper introduces residual module on the basis of denoising convolutional neural network (DnCNN) to strengthen the network learning of low feature information and proposes the feedforward denoising residual network (DnResNet). This study presents that DnResNet achieves better denoising performance than DnCNN using ReLU or LReLU activate function on Set12 and BSD68 dataset with or without batch normalization (BN). DnResNet is superior to DnCNN in average and overall image denoising performance. Therefore, DnResNet is of great significance to image denoising.