Information Theory Oriented Image Restoration

Abstract : This thesis addresses informational formulation of image processing problems. This formulation expresses the solution through a minimization of an information-based energy. These energies belong to the nonparametric class in that they do not make any parametric assumption on the underlying data distribution. Energies are expressed directly as a function of the data considered as random variables. However, classical nonparametric estimation relies on fixed-size kernels which becomes less reliable when dealing with high dimensional data. Actually, recent trends in image processing rely on patch-based approaches which deal with vectors describing local patterns of natural images, e. g., local pixel neighbor- hoods. The k-Nearest Neighbors framework solves these difficulties by locally adapting the data distribution in such high dimensional spaces. Based on these premises, we develop new algorithms tackling mainly two problems of image processing: deconvolution and denoising. The problem of denoising is developed in the additive white Gaussian noise (AWGN) hypothesis and successively adapted to no AWGN realm such as digital photography and SAR despeckling. The denoising scheme is also modified to propose an inpainting algorithm.
Complete list of metadatas
Contributor : Eric Debreuve <>
Submitted on : Thursday, September 8, 2016 - 11:25:43 AM
Last modification on : Monday, November 5, 2018 - 3:52:02 PM
Long-term archiving on : Friday, December 9, 2016 - 12:47:45 PM


  • HAL Id : tel-01362134, version 1



Cesario Vincenzo Angelino. Information Theory Oriented Image Restoration. Signal and Image processing. Université Nice Sophia Antipolis, 2011. English. ⟨tel-01362134⟩



Record views


Files downloads