Abstract:Image denoising is regarded as an ill-posed problem in computer vision tasks that removes additive noise from imaging sensors. Recently, several convolution neural network-based image-denoising methods have achieved remarkable advances. However, it is difficult for a simple denoising network to recover aesthetically pleasing images owing to the complexity of image content. Therefore, this study proposes a multi-branch network to improve the performance of the denoising method. First, the proposed network is designed based on a conventional autoencoder to learn multi-level contextual features from input images. Subsequently, we integrate two modules into the network, including the Pyramid Context Module (PCM) and the Residual Bottleneck Attention Module (RBAM), to extract salient information for the training process. More specifically, PCM is applied at the beginning of the network to enlarge the receptive field and successfully address the loss of global information using dilated convolution. Meanwhile, RBAM is inserted into the middle of the encoder and decoder to eliminate degraded features and reduce undesired artifacts. Finally, extensive experimental results prove the superiority of the proposed method over state-of-the-art deep-learning methods in terms of objective and subjective performances.

What problem does this paper attempt to address?

The paper mainly addresses the problem of image denoising, aiming to improve the performance of existing methods by proposing a new multi-branch network structure. Specifically, the paper tackles the following three main issues: 1. **Simple networks struggle to achieve high-quality denoising results**: Due to the high complexity of image content, simple denoising networks find it difficult to restore satisfactory image quality. 2. **Loss of global information**: When using long-range dependency structures, existing methods often lose global information. 3. **Unwanted artifacts**: Many convolutional neural network (CNN)-based image denoising methods produce unwanted artifacts and lose some critical details because they cannot adapt to changes in texture and edges. To address the above issues, the authors propose a multi-branch network based on the traditional autoencoder structure to learn multi-level contextual features from noisy images. Additionally, the network integrates two modules: the Pyramid Context Module (PCM) and the Residual Bottleneck Attention Module (RBAM) to select important information during training. - **Pyramid Context Module (PCM)**: This module is located at the beginning of the network and uses dilated convolution technology to extract multi-scale contextual features in a pyramid form, thereby expanding the receptive field and effectively addressing the problem of global information loss. - **Residual Bottleneck Attention Module (RBAM)**: This module is inserted in the middle part of the encoder and decoder to focus on valuable features and ignore degraded features, reducing unwanted artifacts. The main contributions of the paper include: - Proposing an effective multi-branch network structure that significantly improves the performance of image denoising tasks. - Introducing the Pyramid Context Module using dilated convolution to expand the receptive field and successfully address the problem of global information loss. - Designing a Residual Bottleneck Attention Module to eliminate degraded features and reduce unwanted artifacts. - Conducting comprehensive experimental validation on multiple datasets, demonstrating that the proposed algorithm outperforms existing deep learning methods in both objective and subjective evaluation metrics. In summary, this paper effectively addresses the challenges present in existing image denoising methods by introducing novel network architectures and modules, thereby enhancing denoising performance.

Multi-Branch Network for Color Image Denoising Using Dilated Convolution and Attention Mechanisms

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

Spatial-Adaptive Network for Single Image Denoising

Dual-constraint Burst Image Denoising Method

Dilated kernel prediction network for

Image Denoising Via Multi-Scale Gated Fusion Network

A Neural Network Based Low-Light Image Denoising Method

Dual Residual Attention Network for Image Denoising

Dilated Residual Encode-Decode Networks for Image Denoising

Dual convolutional neural network with attention for image blind denoising

DCANet: Dual Convolutional Neural Network with Attention for Image Blind Denoising

An efficient lightweight network for image denoising using progressive residual and convolutional attention feature fusion

Two-stage Progressive Residual Dense Attention Network for Image Denoising

Lightweight multi-scale generative adversarial network with attention for image denoising

Residual Dilated Network With Attention For Image Blind Denoising

Color image restoration using DSS-NL-mapping-based multi-noiseNet CNN model

Attention-guided CNN for image denoising

MSPNet: Multi-stage progressive network for image denoising

A Dynamic Network with Transformer for Image Denoising

A Multi-Head Convolutional Neural Network With Multi-path Attention improves Image Denoising