Architectures have been proposed that learn end‐to‐end mapping functions to improve the spatial resolution of natural images. Two arechitectures are proposed. The first one formulates a nonlinear combination of image interpolation techniques using a convolutional neural network. The second architecture uses a skip connection with nearest‐neighbour interpolation, achieving almost similar results. The architectures preserve the sharpness details ls in the image, while also obtaining comparable or better peak‐signal‐to‐noise ratio values than the state‐of‐the‐art, deep learning‐based natural image super‐resolution techniques. The computational complexity of the proposed architectures is much lower than the existing methods. Extensive experiments on five standard datasets for upscale factors of 2 and 4 show that the proposed architectures generalize well.