Abstract:A lot of current work based on convolutional neural networks (CNNs) has fetched good visual results on AWGN (additive white Gaussian noise) removal. However, ordinary neural networks are unable to recover detailed information for complex tasks, and the application of a single Gaussian denoising model is greatly limited. To improve the practicality of the denoising algorithm, we trained a DCNN (deep convolutional neural network) to perform multiple denoising tasks, including Gaussian denoising and blind Gaussian denoising. The proposed CNN denoising model with a residual structure and apply feature attention to exploit channel dependency. The network structure mainly consists of sparse block (SB), feature fusion block (FFB), feature compression block (FCB), information interaction block (IIB) and reconstruction block (RB). The SB with sparse mechanism obtains global and local features by alternating between dilated convolution and common convolution. The FFB collects and fuses global and local features to provide additional information for the latter network. The FCB refines the extracted information and compresses the network. The IIB is used for feature integration and dimensionality reduction. Finally, the RB is used to reconstruct the denoised image. A channel attention mechanism is added to the network, and a trade-off is made between the denoising effect and the complexity of the network. A large number of experiments are conducted on five datasets, and the results showed that the proposed method achieves highly competitive performance in both objective evaluation indicators and subjective visual effects.

Deep Convolutional Neural Network Based on Multi-Scale Feature Extraction for Image Denoising

Grouped Multi-Scale Network for Real-World Image Denoising.

Dynamic Residual Dense Network For Image Denoising

Image Denoising Via Multi-Scale Gated Fusion Network

V-shaped Neural Network Structure Based on Multi-Scale Features for Image Denoising

Texture Compensation with Multi-Scale Dilated Residual Blocks for Image Denoising.

CNN and Multi-Feature Extraction Based Denoising of CT Images

A Multi-scale Dilated Residual Convolution Network for Image Denoising

Multi-Scale Feature Learning Convolutional Neural Network for Image Denoising

A Multiscale Image Denoising Algorithm Based On Dilated Residual Convolution Network

DnM3Net: Multi-Scale & Multi-Level Shuffle-CNN Via Multi-Level Attention for Image Denoising

Multi-scale Dilated Convolution of Convolutional Neural Network for Image Denoising.

Convolutional Neural Network Denoising Method Based on Multisize Features

Image Denoising Via Deep Residual Convolutional Neural Networks

A Multi-Scale Adaptive Feature Enhancement Network for Image Denoising

MMPDNet: Multi-Stage & Multi-Attention Progressive Image Denoising

A Multi-Scale Feature Extraction-Based Normalized Attention Neural Network for Image Denoising

Image Denoising Method Based on a Deep Convolution Neural Network

A Multi-Scale Generative Adversarial Network for Real-World Image Denoising

Multi-scale Network Toward Real-World Image Denoising.

FCNet: a Deep Neural Network Based on Multi-Channel Feature Cascading for Image Denoising