by Cecilia Aguerrebere, Andrés Almansa, Yann Gousseau, Julie Delon, Pablo Musé
Abstract:
Patch models have proven successful to solve a variety of inverse problems in image restoration. Recent methods, combining patch models with a Bayesian approach, achieve state-of-the-art results in several restoration problems. Dif-ferent strategies are followed to determine the patch mod-els, such as a fixed number of models to describe all im-age patches or a locally determined model for each patch. Local model estimation has proven very powerful for im-age denoising, but it becomes seriously ill-posed for other inverse problems such as interpolation of random missing pixels or zooming. In this work, we present a new frame-work for image restoration that combines these two power-ful approaches: Bayesian restoration and a local charac-terization of image patches. By making use of a prior on the model parameters, we overcome the ill-posedness of the local estimation and obtain state-of-the-art results in prob-lems such as interpolation, denoising and zooming. Exper-iments conducted on synthetic and real data show the effec-tiveness of the proposed approach.
Reference:
A Hyperprior Bayesian Approach for Solving Image Inverse Problems (Cecilia Aguerrebere, Andrés Almansa, Yann Gousseau, Julie Delon, Pablo Musé), In (ICCP 2015) IEEE International Conference on Computational Photography, 2015.
Bibtex Entry:
@inproceedings{Aguerrebere2015-HBE-ICCP,
Abstract = {Patch models have proven successful to solve a variety of inverse problems in image restoration. Recent methods, combining patch models with a Bayesian approach, achieve state-of-the-art results in several restoration problems. Dif-ferent strategies are followed to determine the patch mod-els, such as a fixed number of models to describe all im-age patches or a locally determined model for each patch. Local model estimation has proven very powerful for im-age denoising, but it becomes seriously ill-posed for other inverse problems such as interpolation of random missing pixels or zooming. In this work, we present a new frame-work for image restoration that combines these two power-ful approaches: Bayesian restoration and a local charac-terization of image patches. By making use of a prior on the model parameters, we overcome the ill-posedness of the local estimation and obtain state-of-the-art results in prob-lems such as interpolation, denoising and zooming. Exper-iments conducted on synthetic and real data show the effec-tiveness of the proposed approach.},
Address = {Rice University, Houston, TX},
Annote = {poster presentation {\#}44 at ICCP 2015
http://iccp.rice.edu/posters{\_}and{\_}demos/
Longer preprint:
https://hal.archives-ouvertes.fr/hal-01107519},
Author = {Aguerrebere, Cecilia and Almansa, Andr{\'{e}}s and Gousseau, Yann and Delon, Julie and Mus{\'{e}}, Pablo},
Booktitle = {(ICCP 2015) IEEE International Conference on Computational Photography},
Keywords = {Bayesian restoration,Gaussian Mixture Models,Maximum a Posteriori,Non-local patch-based restoration,conjugate distributions,hyper-prior},
Language = {en},
Title = {{A Hyperprior Bayesian Approach for Solving Image Inverse Problems}},
Url = {https://hal.archives-ouvertes.fr/hal-01107519v1},
Year = {2015},
Bdsk-Url-1 = {https://hal.archives-ouvertes.fr/hal-01107519v1}}