Deep learning methods to enhance adventure films

Details

Supervisors

Dupont, Pierre

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en informatique, à finalité spécialisée

Abstract

In recent years, the fields of computer vision and image processing have been revolutionized by machine learning techniques and in particular by deep learning models. Indeed, a lot of image processing methods have been rediscovered, even surpassed, by these new artificial intelligence models. The technique that this article will focus on is super-resolution. Its purpose is to increase the number of pixels in an image while improving its visual quality. Whether it is to improve a single image or a video that is a succession of images, super-resolution models have achieved remarkable results. This work will attempt to use the good methods behind this technology for a restoration task. That is, the aim will no longer be to increase the resolution, but only to improve the visual quality of the images. In particular, the input images will be spatially reduced within the model to recover the original dimensions in the output after the super-resolution operation. With this design, the majority of the computations are performed on low-resolution inputs and at the same time allows to improve performance.