We assess the benefit of including an image inpainting filter before passing\ndamaged images into a classification neural network. For this we employ a\nmodified Cahn-Hilliard equation as an image inpainting filter, which is solved\nvia a finite volume scheme with reduced computational cost and adequate\nproperties for energy stability and boundedness. The benchmark dataset employed\nhere is MNIST, which consists of binary images of handwritten digits and is a\nstandard dataset to validate image-processing methodologies. We train a neural\nnetwork based of dense layers with the training set of MNIST, and subsequently\nwe contaminate the test set with damage of different types and intensities. We\nthen compare the prediction accuracy of the neural network with and without\napplying the Cahn-Hilliard filter to the damaged images test. Our results\nquantify the significant improvement of damaged-image prediction due to\napplying the Cahn-Hilliard filter, which for specific damages can increase up\nto 50% and is in general advantageous for low to moderate damage.\n