PROJECT TITLE:

Exemplar-Based Inpainting Technical Review and New Heuristics for Better Geometric Reconstructions - 2015

ABSTRACT:

This paper proposes a technical review of exemplar-primarily based inpainting approaches with a particular concentrate on greedy ways. Many comparative and illustrative experiments are provided to deeply explore and enlighten these ways, and to have a higher understanding on the state-of-the-art enhancements of those approaches. From this analysis, three enhancements over Criminisi et al. algorithm are then presented and detailed: one) a tensor-based mostly knowledge term for a higher choice of pixel candidates to fill in; a pair of) a fast patch lookup strategy to ensure a higher global coherence of the reconstruction; and three) a unique quick anisotropic spatial mixing algorithm that reduces typical block artifacts using tensor models. Relevant comparisons with the state-of-the-art inpainting strategies are provided that exhibit the effectiveness of our contributions.