Single Image Super-resolution Reconstruction Using Convolutional Neural Networks

To reconstruct high resolution image with smooth edges, a single image super-resolution algorithm on convolutional neural networks is proposed in this paper. Very small receptive fields were adapted throughout the whole net, extracting gradient information effectively. By presenting a model with 6 weight layers, we obtained high resolution images with smoother edges and suppressed ringings to a certain extent. Also a larger training set was used in proposed algorithm to avoid over-fitting. While proposed algorithm has little improvement on training set given by Dong’s SRCNN algorithm, it achieves a better performance both in subjective and objective quality evaluation on a relatively larger training set extracted from Image Net.