Super-Resolution Image Reconstruction based on Random-coupled Neural Network and EDSR

Super-resolution image reconstruction is crucial for numerous applications, including chip manufacturing, medical imaging, and vision technology. Although the Enhanced Deep Super-Resolution Network (EDSR) delivers satisfactory outcomes, it struggles with limited multi-scale feature representation, causing blurred edges and texture degradation. In this paper, we introduce the EDSR-RCNN model, a novel approach integrating the Random-Coupled Neural Network (RCNN) within the EDSR framework to effectively enhance mid-to-high-scale feature extraction. The RCNN generates global ignition maps without extensive pre-training, capturing critical high-frequency details such as edges and textures. Quantitative experiments using the DIV2K dataset indicate that EDSR-RCNN surpasses existing methods, achieving superior results in PSNR, SSIM, FSIM, and LPIPS metrics. Qualitative analyses further validate that the model significantly reduces edge blur and artifacts, leading to enhanced visual clarity and fidelity. The proposed method is publicly accessible at https://github.com/HaoranLiu507/EDSR-RCNN.