Single image super resolution

Abstract

In the field of computer vision, super resolution with deep learning is a promising field that has generated multiple research, and has seen its application far and wide. In single image super resolution, the image can either be upscaled before being input into the network (pre upscaling) and its features learned, or it can be upscaled after the network has learned its feature (post upscaling). As with any deep learning models, the inputs into the model can affect the outputs produced. Hence, the goal of this project is to find out how much we can improve a super resolution model by filtering the inputs using a classification model. In this report, we will be discussing our analysis of the data, methodology and models used for classification/super resolution, and the results produced from our experiments. We will also be discussing some of the limitations of the project and future work regarding this project.