Seol mar théacs é seo: Enhancement of damaged-image prediction through Cahn–Hilliard image inpainting