Skip to Main content Skip to Navigation
Master Thesis

Restauration d’image via une approche Bayésienne à partir d’un couple d’images d’observations prise dans différentes conditions

Abstract : Through this memory engineering, the problem of "Computational Photography" is discussed. The goal is to overcome the limits of a traditional camera using informatics techniques to improve the visual quality of the image produced. Two solutions are studied: deconvolution with or without regularization, and blind deconvolution. Despite the benefits of these techniques, deconvolution criteria also present important drawbacks, such as the existence of the inverse problem, the presence of local extrema associated with parasites solutions, and system instability. So too many restoration methods are used in order to overcome the shortcomings of the damage, where the target is to contain an image like the original scene to can analyze. In our experiments, we place in the context of a camera embarks on a car or a helicopter drone. It goes without saying that these images must be of good quality to be usable. But such equipment mounted on aircraft is subject to much riskiness: climate, engine vibration, moving the camera during the acquisition, etc... Degradation resulting of blurring and noise are sometimes seen on the images obtained, reducing their quality and making it more difficult to extract suitable information. In this framework, we present a new restoration method that uses a pair of images of a single observation as the pair of images is taken with different exposure times. Our contribution is to establish a method of blind déconvolution (Blurring kernel unknown) for the construction of high-quality images by combining two images taken under poor qualities, e.g. taken under different conditions: first blurred (slow shutter speed image) and the second noise ( fast shutter speed image). We formulate the problem in a hierarchical Bayesian framework by using a prior knowledge on the unknown image and blur, and also on the dependency between the two observed images. By incorporating a fully Bayesian analysis, the developed algorithm estimates all necessary model parameters along with the unknown image and blur, such that no user- intervention is needed or required. Moreover, we use a variational Bayesian inference procedure, which allows for the statistical compensation of errors occurring at different stages of the restoration, and also provides uncertainties of the estimates. Experimental results with synthetic and real images demonstrate that the proposed method provides very high quality restoration results and compares favorably to existing methods even though no user supervision is needed.
Document type :
Master Thesis
Complete list of metadatas
Contributor : Cnam - Service Commun de la Documentation - Bibliothèque Centrale <>
Submitted on : Wednesday, November 9, 2016 - 9:36:31 AM
Last modification on : Saturday, February 8, 2020 - 12:02:02 PM
Long-term archiving on: : Tuesday, March 14, 2017 - 11:57:42 PM


  • HAL Id : dumas-01394266, version 1



Mouhamad Chéhaitly. Restauration d’image via une approche Bayésienne à partir d’un couple d’images d’observations prise dans différentes conditions. Electronique. 2013. ⟨dumas-01394266⟩



Record views


Files downloads