Example maps illustrating the computation of the regularization weights for Eq. (14), to properly scale the multiscale regularization of the pixelated lens potential model ψ_pix. Top-left panel: noise standard deviation estimated from a preliminary smooth model of the source, that is the square root of the diagonal elements of C_d, without the contribution from the lens light. The remaining panels show the noise standard deviation propagated to each potential pixel for the first four wavelet scales of the starlet transform, that we used as the regularization weights W_ψ, st. We see that the standard deviation strongly decreases as a function of the wavelet scale. The resulting maps are similar for the Battle-Lemarié wavelet transform. More details are given Sect. 2.4.