Title:Optimal control of the fidelity coefficient in a Cahn-Hilliard image inpainting model

Abstract:We consider an inpainting model proposed by A. Bertozzi et al., which is based on a Cahn--Hilliard-type equation. This equation describes the evolution of an order parameter $u \in [0,1]$ representing an approximation of the original image which occupies a bounded two-dimensional domain $\Omega$. The given image $g$ is assumed to be damaged in a fixed subdomain $D \subset \Omega$ and the equation is characterized by a linear reaction term of the form $\lambda (u - g)$. Here $\lambda = \lambda_0 \chi_{\Omega \setminus D}$ is the so-called fidelity coefficient, $\lambda_0$ being a strictly positive bounded function. The idea is that, given an initial image $u_0$, $u$ evolves towards $g$ and this process properly diffuses through the boundary of $D$ restoring the damaged image, provided that $\lambda_0$ is large enough. Here, we formulate an optimal control problem based on this fact, namely our cost functional accounts for the magnitude of $\lambda_0$. Assuming a singular potential to assure that $u$ takes its values in $[0,1]$, we first analyse the control-to-state operator and prove the existence of at least one optimal control, establishing the validity of first-order optimality conditions. Then, under suitable assumptions, we demonstrate second-order optimality conditions. All these results depend on the existence and uniqueness of a strong solution, which we obtain thanks to the strict separation property from pure phases.