Image deblurring based on local features and long-range dependencies
Jiahao Wang, Qing Qi, Kaiqiang Zhang · 2024
Dynamic scene deblurring is a complex problem that involves blurring caused by camera shake and object movement. While deep learning methods have made significant progress in many the field of image deblurring, there is still room for improvement. This study aims to develop a deblurring model based on generative adversarial networks, which integrates Multi-Scale Residual Dense Network (MSRDN) and attention mechanisms. MSRDN focus on capturing local features of images, ensuring that the model can grasp the detailed information of images. Meanwhile, the attention mechanism is responsible for learning long-range dependencies in images, further enhancing the model’s ability to grasp global information. The combination of these dual mechanisms significantly improves the model’s performance in capturing and learning deep features of images. Experimental results demonstrate that our approach can generate photo-realistic clear images from blurred images in real-world scenarios, outperforming existing state-of-the-art methods.