Lineární algebra rekonstrukce obrazu

Abstract

A mathematical representation of digital image with particular interest in blurring formulation is described from linear algebra point of view. Arising minimization problem is then solved by various quasi-Newton methods for unconstrained minimization (Landweber and RNSD) and nonnegatively constrained minimization, where the nonnegativity is treated by gradient projection (FSGP and RNSD with projection) or parametrization (MRNSD). The convergence of the methods is tested by numerical experiments on sample image.