Fusion Objective Function on Progressive Super-Resolution Network

Recent advancements in Single-Image Super-Resolution (SISR) have explored the network architecture of deep-learning models to achieve a better perceptual quality super-resolved images. However, effect objective function, which contributes improving performance and images, has not gained much attention. This paper proposes novel super-resolution called Progressive Multi-Residual Fusion Network (PMRF), fuses learning functions L2 Multi-Scale SSIM progressively upsampling framework structure. Specifically, we propose Residual-in-Residual Dense Blocks (RRDB) on platform that reconstructs high-resolution image during intermediate steps our network. Additionally, Depth-Wise Bottleneck Projection allows high-frequency information early layers be bypassed through modules Quantitative qualitative evaluation benchmark datasets demonstrate proposed PMRF algorithm with fusion function (L2 MS-SSIM) improves model’s accuracy compared other state-of-the-art models. Moreover, this model demonstrates robustness against noise degradation achieves an acceptable trade-off between efficiency accuracy.