Single image super resolution using deep convolutional generative neural networks

Abstract

Nowadays, deep convolutional networks have been focused on single image super-resolution problem due to their impressive performance on generating high-resolution images like as other computer vision tasks. It is clearly seen that among best known super-resolution models deep learning-based methods give the-state-of-the-art results. In this study, FSRGAN, based on a popular deep convolutional network (FSRCNN) due to its efficiency in spite of its simple architecture, is presented with generative adversarial training approach combining a discriminative network to the generator. The performance of the presented model is demonstrated by comparing to its baseline model, which is used as a generative network of our FSRGAN, the interpolation methods on well-known data sets based on PSNR metric.