Evaluation of digital inpainting quality in the context of artwork restoration

Authors:
Alexandra Ioana Oncu;Ferdinand Deger;Jon Yngve Hardeberg
Affiliations:
The Norwegian Colour and Visual Computing Laboratory, Gjøvik University College, Gjøvik, Norway;The Norwegian Colour and Visual Computing Laboratory, Gjøvik University College, Gjøvik, Norway;The Norwegian Colour and Visual Computing Laboratory, Gjøvik University College, Gjøvik, Norway
Venue:
ECCV'12 Proceedings of the 12th international conference on Computer Vision - Volume Part I
Year:
2012

Citing 7
Cited 0

Vector-Valued Image Regularization with PDEs: A Common Framework for Different Applications

IEEE Transactions on Pattern Analysis and Machine Intelligence
Fast Image Inpainting Based on Coherence Transport

Journal of Mathematical Imaging and Vision
PatchMatch: a randomized correspondence algorithm for structural image editing

ACM SIGGRAPH 2009 papers
A New Spatial Hue Angle Metric for Perceptual Image Difference

Computational Color Imaging
Image Inpainting Based on Local Optimisation

ICPR '10 Proceedings of the 2010 20th International Conference on Pattern Recognition
Region filling and object removal by exemplar-based image inpainting

IEEE Transactions on Image Processing
VSNR: A Wavelet-Based Visual Signal-to-Noise Ratio for Natural Images

IEEE Transactions on Image Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

Improved digital image inpainting algorithms could provide substantial support for future artwork restoration. However, currently, there is an acknowledged lack of quantitative metrics for image inpainting evaluation. In this paper the performance of eight inpainting algorithms is first evaluated by means of a psychophysical experiment. The ranking of the algorithms thus obtained confirms that exemplar based methods generally outperform PDE based methods. Two novel inpainting quality metrics, proposed in this paper, eight general image quality metrics and four inpainting-specific metrics are then evaluated by validation against the perceptual data. Results show that no metric can adequately predict inpainting quality over the entire image database, and that the performance of the metrics is image-dependent.