Wavelet compression is a well-settled technique in image pipelines (such as satellite imagery and JPEG2000) and the quantification of noisy coefficients often causes outliers, which result in highly-correlated artifacts in the spatial domain. We propose a joint restoration scheme that uses a precise degradation model taking into account the intertwined effects of noise and compression. Different kinds of regularization, including a novel Bayesian patch-based approach, Gaussian Mixture Models and Convolutional Neural Networks are explored.

This is poster number 53 in Poster Session

Authors:

Mario Gonzalez (Universidad de la República (Uruguay), Université Paris Descartes)

Andrés Almansa (MAP5 - CNRS - Université Paris Descartes)

Pablo Musé (Facultad de Ingeniería, Universidad de la República)

Keywords: