Abstract:Image restoration is a long-standing task that seeks to recover the latent sharp image from its deteriorated counterpart. Due to the robust capacity of self-attention to capture long-range dependencies, transformer-based methods or some attention-based convolutional neural networks have demonstrated promising results on many image restoration tasks in recent years. However, existing attention modules encounters limited receptive fields or abundant parameters. In order to integrate contextual information more effectively and efficiently, in this paper, we propose a dilated strip attention network (DSAN) for image restoration. Specifically, to gather more contextual information for each pixel from its neighboring pixels in the same row or column, a dilated strip attention (DSA) mechanism is elaborately proposed. By employing the DSA operation horizontally and vertically, each location can harvest the contextual information from a much wider region. In addition, we utilize multi-scale receptive fields across different feature groups in DSA to improve representation learning. Extensive experiments show that our DSAN outperforms state-of-the-art algorithms on several image restoration tasks.

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve This paper aims to address the critical issue in image restoration. Image restoration is a long-standing task that aims to recover a clear original image from a degraded one. Specifically, this task involves recovering high-quality images from those affected by blurring, snowflakes, or haze. Although existing transformer models based on self-attention mechanisms perform well in capturing long-range dependencies, they face challenges in handling high-resolution images due to high computational complexity. ### Main Issues and Solutions 1. **Limitations of Existing Methods**: - **Limitations of Convolutional Neural Networks (CNNs)**: Traditional convolutional neural networks struggle to handle dynamic and non-uniform blurring issues due to their static filters and limited receptive fields. - **Limitations of Self-Attention Mechanisms**: While self-attention mechanisms can effectively model long-range dependencies, they suffer from high computational complexity when processing high-resolution images, leading to inefficiency. 2. **Proposed Methods**: - **Dilated Strip Attention (DSA)**: To more effectively integrate contextual information, the authors designed a dilated strip attention mechanism. By applying DSA operations in both horizontal and vertical directions, each pixel can obtain contextual information from a broader area. - **Dilated Strip Attention Module (DSAM)**: The DSAM is further proposed to sequentially integrate information effectively in horizontal and vertical directions, significantly expanding the receptive field and capturing more contextual information. - **Dilated Strip Attention Network (DSAN)**: By integrating DSAM into a U-shaped network structure, the DSAN is constructed for efficient and effective image restoration. ### Experimental Results - **Image Dehazing**: Experimental results on the SOTS dataset show that DSAN achieved the highest PSNR and SSIM values in both indoor and outdoor scenes, reaching 40.60 dB and 38.41 dB, respectively, outperforming other state-of-the-art methods. - **Image Motion Deblurring**: Experimental results on the GoPro and HIDE datasets show that DSAN outperformed powerful transformer models like Restormer in terms of PSNR and SSIM metrics. - **Image Desnowing**: Experimental results on the CSD dataset show that DSAN significantly outperformed other state-of-the-art methods in terms of PSNR and SSIM metrics, reaching 36.56 dB and 0.985, respectively. ### Conclusion By introducing the dilated strip attention mechanism and module, DSAN performs excellently in multiple image restoration tasks, not only surpassing existing methods in performance but also having advantages in computational efficiency. This makes DSAN a significant advancement in the field of image restoration.

Dilated Strip Attention Network for Image Restoration

Dual-domain strip attention for image restoration

Image Restoration Using Dual-Domain Fusion Network for Rotating Rectangular Synthetic Aperture System

Local to non-local: Multi-scale progressive attention network for image restoration

Accurate Image Restoration with Attention Retractable Transformer

How Powerful Potential of Attention on Image Restoration?

Learning Deformable and Attentive Network for image restoration

A Double Feature Fusion Network with Progressive Learning for Sharper Inpainting

Residual Non-local Attention Networks for Image Restoration

Image Restoration Via Deep Memory-Based Latent Attention Network

Emphasizing Crucial Features for Efficient Image Restoration

HAT: Hybrid Attention Transformer for Image Restoration

Dilated Residual Encode-Decode Networks for Image Denoising

Single Stage Adaptive Multi-Attention Network for Image Restoration

Empowering Image Recovery_ A Multi-Attention Approach

Synergic feature attention for image restoration

Towards Context-aware Convolutional Network for Image Restoration

SCGA‐Net: Skip Connections Global Attention Network for Image Restoration

Joint multi-dimensional dynamic attention and transformer for general image restoration

Focal Network for Image Restoration

DRCNet: Dynamic Image Restoration Contrastive Network.