告题目:Multigrid iterative regularization with framelet denoising for image deblurring
报告人:Marco Donatelli(意大利因苏布里亚大学副教授)
报告时间:2018年4月19日下午15:30
报告地点:清水河校区 主楼A1 – 512
报告摘要:Recent numerical
methods for image deblurring decompose nonlinear problems into simpler
subproblems and usually require an inner step of linear regularization. For
instance, the classical iterative shrinkage thresholding algorithm combines one
step of Landweber method with one step of soft-thresholding. Assuming that an
estimation of the noise level is available, we combine a recently proposed
iterative regularization method with linear framelet denoising. This
combination allows a fast frequency filtering and produces a sparse
reconstruction in the wavelet domain. Moreover, its employment in a multigrid
scheme ensures a stable convergence and a reduced noise amplification. The
coarsening strategy used in our multigrid method is defined by the low-pass
filter used for the framelet denoising and is able to achieve fast computations
at each coarser level independently of the imposed boundary conditions.
Finally, the iterations are projected in a closed and convex set that is known
to contain the exact solution of the problem, e.g., the nonnegative cone. We study
the convergence of the proposed algorithm and prove that, if supplied with the
suitable stopping criterion, it is a regularization method. The comparison with
other methods from the literature proves that it is able to provide highly
accurate reconstruction without requiring the setting of any parameter.
报告人简介: Prof. Marco Donatelli是意大利因苏布里亚大学科学与高科技系-数值分析组的副教授。目前担任SIAM J. Sci. Comput., SIAM J. Imaging Sci., SIAM J. Matrix Anal. Appl., IMA J. Numer. Anal., Inverse Problems, Appl. Math. Comput., Linear Algebra Appl., BIT, J. Comput. Appl. Math., Appl. Numer. Math.等著名国际杂志编委。他还是National Group for Scientific Calcolus (GNCS),Italian Society of Applied and Industrial Mathematics (SIMAI),Society for Industrial and Applied Mathematics (SIAM) 的成员。