Multi-Path Attention Block-Loss Function for Non-Uniform Blind Image Deblurring

Blind image deblurring is a significant and difficult task in the computer vision field. It is mainly focused on decreasing the blur caused by motion or camera shake. However, image deblurring has made great progress, but there is a need for enhancement in the visual effect and image information to enhance the overall image quality. In this research, the Multi-Path Attention Block-Loss Function (MPAB-LF) is proposed for non-uniform blind image deblurring to enhance image quality. The MPAB approach is employed to provide the foundation for blind image deblurring. To combine numerous MPABs instead of employing associated feature aggregation, the Improved One-Shot Aggregation (IOSA) is utilized. Finally, the loss functions are employed to enhance the generator and discriminator of the framework. Existing approaches like Generative Adversarial Network (GAN), MPAB, Two Convolutional Neural Networks (CNN), and Dense Dilated Block and Improved Attention Module (DDBIAM) are used to compare with the proposed MPAB-FL approach. The proposed MPAB-FL achieves a better Peak-to-Signal Noise Ratio (PSNR) of 35.65 dB and Structural Similarity Index (SSIM) of 0.9957 compared to the existing approaches like GNN, MPAB, Two-stage CNN, and DDBIAM respectively.