Super‐resolution using deep residual network with spectral normalization

In this letter, the authors present a single-image super-resolution method based on introducing novel spectral normalization to convolution of deep residual network. Moreover, construct new block (RB) and assemble it in cascade form. The RB was restructured by spectrally normalized layers activated function. addition, allows introduced trained network update additional weights. Furthermore, minimizes pixel loss, thereby helping obtain enhanced reconstruction results, limiting number parameters, facilitating low computational cost. experimental results demonstrate that proposed model shows superior performance over state-of-the-art methods terms visual quality metrics such as UQI PIQE.