“…Hence, we can alternate a minimisation step with respect to the primal variables t, u, w with a maximisation step with respect to the dual variables ρ t , ρ w , in combination with an iterative update of the space variant parameters α i and µ, which hence will be denoted by α we use the easy ML estimation strategy described next, whereas for µ (k) we will rely on a global discrepancy principle Primal variables update. The three primal sub-problems can be solved efficiently and in closedform by simple shrinkage/projection operators (in both cases p = 1, 2) and linear system solvers -see [13,12]. More in details, the sub-problem with respect to the t primal variable, after some algebraic manipulations, can be written as,…”