Transformer-based progressive residual network for single image dehazing.

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Degraded fog images hinder computer vision tasks.
• Obtaining clear images from foggy conditions is a significant challenge.
• Vision Transformer (ViT) architectures show promise in various vision applications.

Purpose of the Study:

• To develop an effective transformer-based method for single image de-fogging.
• To improve the performance of subsequent visual tasks by generating fog-free images.
• To introduce a novel progressive residual network incorporating Swin Transformer.

Main Methods:

• A progressive residual network is proposed, recursively utilizing Swin Transformer blocks.
• The network comprises recurrent blocks, transformer codecs, and a supervise fusion module.
• Attention mechanisms, including channel attention, are integrated for feature fusion and selection.

Main Results:

• The proposed method demonstrates superior performance compared to state-of-the-art approaches.
• Experimental results validate the effectiveness of the transformer-based de-fogging technique.
• The approach successfully generates fog-free images, enhancing visual quality.

Conclusions:

• The novel transformer-based progressive residual network offers a powerful solution for image de-fogging.
• This method achieves state-of-the-art results, outperforming existing handcrafted and learning-based techniques.
• The approach has significant implications for improving computer vision applications reliant on clear imagery.