Deep learning for image restoration

The activity involves the development of deep learning architectures for several image restoration problems, including image denoising, single-image and multi-image super-resolution, image deblurring/deconvolution, and reconstruction from projections. We have employed several types of deep neural networks for these problems, including CNNs, graph neural networks, generative adversarial networks, obtaining state-of-the-art results via the custom design of highly efficient architectures based on permutation invariance.

Application areas are in the field of satellite imaging and computational imaging for smartphones and cameras.


ERC Sector:

  • PE7_7 Signal processing


  • Artificial neural networks
  • Image processing
  • Deep learning
  • Image restoration