Summary
IEICE Information and Communication Technology Forum
2013
Session Number:ENG
Session:
Number:ENG-2
Context Aware Image Inpainting with Application to Virtual Restoration of Old Paintings
Tijana Ruzic,
pp.-
Publication Date:2013-07-20
Online ISSN:2188-5079
Summary:
In this paper, we explore how to use spatial context for image inpainting and we test it in two applications: photo-editing and crack removal in digitized old paintings. Context is determined based on the texture and color features. We use these contextual features to guide the search for image patches that can fill in the missing/damaged regions in a visually plausible way. A priori knowledge about spatial consistencies (similarities) among neighbouring image patches is encoded via a Markov Random Field (MRF) model. With this prior, the process of image inpainting is formulated as an optimization problem. We define an efficient inference engine as a valuable alternative to the so-called belief propagation methods. As a case study, we focus on crack removal from the Adoration of the Mystic Lamb (brothers Van Eyck, 1432), demonstrating potentials for virtual restoration of old paintings.