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

José A Carrillo,Sergio P Perez,Fuyue Liang,Serafim Kalliadasis

doi:10.1098/rsos.201294

Abstract

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

Highlights

Image inpainting consists in filling damaged or missing areas of an image, with the ultimate objective of restoring it and making it appear as the true and original image
We have quantified the prediction improvement of employing a CH image inpainting filter to restore damaged images which are passed into a neural network
We combined a finite-volume scheme with a neural network for pattern recognition to develop an integrated algorithm summing up the process of adding damage to the images and predicting their label

Summary

Introduction

Image inpainting consists in filling damaged or missing areas of an image, with the ultimate objective of restoring it and making it appear as the true and original image. Soc. Open Sci. 8: 201294 professional restorers, but it was not until the turn of the twenty-first century that digital image inpainting 2 models based on partial-differential equations (PDEs) and variational methods were introduced [1,2,3]. Open Sci. 8: 201294 professional restorers, but it was not until the turn of the twenty-first century that digital image inpainting 2 models based on partial-differential equations (PDEs) and variational methods were introduced [1,2,3] These methods are usually referred to as non-texture, geometrical or structural inpainting since they focus on restoring the structural information in the inpainted domain such as edges, corners or curvatures. We focus on non-texture image inpainting methods based on partial differential equations (PDEs), and we refer the reader to [10] for a general review of the topic

Objectives

Methods

Findings

Conclusion