Abstract: Extensively, image super-resolution (SR) poses a challenge across all fields of interest as its problem is considered inherent in its acquisition due to several reasons. Hence, many algorithms have been proposed to suppress this inherent challenges. As a contribution to help see through this inherency, we modelled a Randomized Convolutional Neural Network for Image Super-Resolution (RCNNSR) which simply learns an end-to-end mapping existing between the low-resolution (LR) and the high-resolution (HR) and this reconstructed high-resolution image is kindred as possible with the corresponding ground truth high-resolution image.