Navigating the Complex Landscape of Shock Filter Cahn–Hilliard Equation: From Regularized to Entropy Solutions

Image inpainting involves filling in damaged or missing regions of an image by utilizing information from the surrounding areas. In this paper, we investigate a fully nonlinear partial differential equation inspired by the modified Cahn–Hilliard equation. Instead of using standard potentials that depend solely on pixel intensities, we consider morphological image enhancement filters that are based on a variant of the shock filter:

$$\begin{aligned} \partial _t u&= \Delta \left( -\nu \arctan (\Delta u)|\nabla u| - \mu \Delta u \right) + \lambda (u_0 - u). \end{aligned}$$

This is referred to as the Shock Filter Cahn–Hilliard Equation. The equation is nonlinear with respect to the highest-order derivative, which poses significant mathematical challenges. To address these, we make use of a specific approximation argument, establishing the existence of a family of approximate solutions through the Leray–Schauder fixed point theorem and the Aubin–Lions lemma. In the limit, we obtain a solution strategy wherein we can prove the existence and uniqueness of solutions. Proving the latter involves the Kruzhkov entropy type-admissibility conditions. Additionally, we use a numerical method based on the convexity splitting idea to approximate solutions of the nonlinear partial differential equation and achieve fast inpainting results. To demonstrate the effectiveness of our approach, we apply our method to standard binary images and compare it with variations of the Cahn–Hilliard equation commonly used in the field.