Super Resolution for remotely sensed imagery

by Savvas Karatsiolis (SuPerWorld Research Group)

  • Theatro Polis OPAP
Savvas Karatsiolis

Remotely sensed images are generally noisy and of low resolution which makes the development of computer vision applications very hard. State-of-the-art Super Resolution algorithms do not perform well on remote-sensing images and tend to create artifacts and distortion on large texture surfaces. We present a novel algorithm that uses information from Digital Surface Models to upsample low resolution images. The inferred reconstructions have x4 or x8 higher resolution than the input image and the results are often indistinguishable from the ground truth images.

Bio

Dr. Savvas Karatsiolis received his Ph.D. degree in Machine Learning from University of Cyprus, in 2019 with a focus on Deep learning. His area of research includes machine learning theory, computer vision, generative models, unsupervised and self-supervised learning.