Résumé/Abstract :
“A picture is worth a thousand words”. However, in a blurred image, most information of image is hidden. To reveal the hidden information, we need a blind deblurring algorithm.

Blind deblurring is not an easy task because we must estimate two variables, the blurring function and the original image, from only one equation, the blurred observation model. In addition, the small noise in blurred observation would be amplified significantly in the restored image as deblurring is an ill-posed inverse problem.

This seminar will present a new approach to solving the deblurring problem. We will estimate the clear image of the real scene from only one noisy blurred image of this scene. In our context, the blurring phenomenon is characterized by a spatially invariant blurring function and the contaminated noise is an additive white Gaussian random process. The proposed algorithms use the Variational Bayesian approach and a new adaptive image model.