Enhancement of damaged-image prediction through Cahn–Hilliard image inpainting

We assess the benefit of including an image inpainting filter before passing damaged images into a classification neural network. employ appropriately modified Cahn–Hilliard equation as which is solved numerically with finite-volume scheme exhibiting reduced computational cost and properties energy stability boundedness. The benchmark dataset employed Modified National Institute Standards Technology (MNIST) dataset, consists binary handwritten digits standard to validate image-processing methodologies. train network based on dense layers MNIST, subsequently we contaminate test set damages different types intensities. then compare prediction accuracy without applying test. Our results quantify significant improvement damaged-image by filter, for specific can increase up 50% advantageous low moderate damage.