Optimized Lightweight Network for Enhanced Image Denoising

Image denoising plays a vital role in enhancing image quality by effectively removing noise while preserving essential details, a crucial step for numerous computer vision applications. In this paper, we propose an optimized lightweight network for enhanced image denoising that strikes a balance between high denoising performance and computational efficiency. The proposed model utilizes a novel architecture based on Bi-Conv Blocks, which incorporate both standard and dilated convolutions to capture multi-scale features efficiently. This dual-pathway design, combined with local and global residual connections, enables the model to preserve fine-grained details and global image structures while effectively removing noise across varying levels. The local residual connections facilitate better feature learning within each Bi-Conv Block, while the global residual connection ensures the retention of the global context throughout the network. To further optimize the model for real-world use, a patch-based processing approach is adopted, allowing efficient denoising of localized regions and reducing both memory usage and computational cost. Extensive experiments on benchmark datasets demonstrate that our model outperforms existing state-of-the-art methods in terms of Peak Signal-to-Noise Ratio (PSNR) and structural similarity, particularly in challenging high-noise environments. The lightweight architecture of the proposed model makes it highly suitable for real-time applications and deployment in resource-constrained environments.