An End-to-End Framework for Image Super Resolution and Denoising of SAR Images

Single image super resolution (or upscaling) has become very efficient because of the powerful application of generative adversarial networks (GANs). However, the presence of noise in the input image often produces undesired artifacts in the resultant output image. Denoising an image and then upscaling introduces more chances of these artifacts due to the accumulation of errors in the prediction. In this work, we propose a single shot upscaling and denoising of SAR images using GANs. We have compared the quality of the output image with the two-step denoising and upscaling network. To evaluate our standing with respect to the state-of-the-art, we compare our results with other denoising methods without super resolution. We also present a detailed analysis of experimental findings on the publicly available COWC dataset, which come with context information for object classification.