Super-Resolution Image using Enhanced Deep Residual Networks and the DIV2K Dataset
S. Senthilkumar, S. Vetriselvi, K. S. Kalaivani, P Arunkumar, M. Malathi, P. V. Praveen Sundar · 2024
Recent advances in deep learning have significantly improved the performance of super-resolution (SR) techniques, yet challenges persist in balancing image quality and computational efficiency. This paper presents a novel approach utilizing Enhanced Deep Residual Networks (EDRN) to effectively address these challenges. Utilizing the high-quality and diverse DIV2K dataset, this method employs a combination of Mean Squared Error (MSE) and perceptual loss functions to achieve an optimal balance between pixel accuracy and perceptual quality. Comprehensive evaluation using metrics like Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) demonstrates that the proposed EDRN consistently outperforms baseline models, such as SRCNN, VDSR, and EDSR, by delivering superior detail restoration and minimizing patterns. Visual comparisons further indicate the enhanced quality of the super-resolved images, making this approach a valuable asset for high-resolution imaging applications.