This paper focuses on a machine learning approach for objective inpainting quality assessment. Inpainting has received a lot of attention in recent years and quality assessment is an important task to evaluate different image reconstruction approaches. Quantitative metrics for successful image inpainting currently do not exist; researchers instead are relying upon qualitative human comparisons in order to evaluate their methodologies and techniques. We present an approach for objective inpainting quality assessment based on natural image statistics and machine learning techniques. Our method is based on observation that when images are properly normalized or transferred to a transform domain, local descriptors can be modeled by some parametric distributions. The shapes of these distributions are different for noninpainted and inpainted images. Approach permits to obtain a feature vector strongly correlated with a subjective image perception by a human visual system. Next, we use a support vector regression learned on assessed by human images to predict perceived quality of inpainted images. We demonstrate how our predicted quality value repeatably correlates with a qualitative opinion in a human observer study.

Date Published: 22 May 2014
PDF: 9 pages
Proc. SPIE 9120, Mobile Multimedia/Image Processing, Security, and Applications 2014, 91200S (22 May 2014); doi: 10.1117/12.2063664

Published in SPIE Proceedings Vol. 9120:
Mobile Multimedia/Image Processing, Security, and Applications 2014
Sos S. Agaian; Sabah A. Jassim; Eliza Yingzi Du, Editor(s)