By Topic

Turbo inpainting: Iterative K-SVD with a new dictionary

Sign In

Cookies must be enabled to login.After enabling cookies , please use refresh or reload or ctrl+f5 on the browser for the login options.

Formats Non-Member Member
$33 $13
Learn how you can qualify for the best price for this item!
Become an IEEE Member or Subscribe to
IEEE Xplore for exclusive pricing!
close button

puzzle piece

IEEE membership options for an individual and IEEE Xplore subscriptions for an organization offer the most affordable access to essential journal articles, conference papers, standards, eBooks, and eLearning courses.

Learn more about:

IEEE membership

IEEE Xplore subscriptions

2 Author(s)
Min-Sung Koh ; School of Computing and Engineering Science, Eastern Washington University, Cheney, 99004, USA ; Esteban Rodriguez-Marek

This paper introduces a new inpainting technique denominated turbo inpainting. The algorithm modifies the K-clustering with singular value decomposition (K-SVD) inpainting method to allow for iterative, progressive improvement. Turbo inpainting is composed of two separate blocks: inner inpainting and outer inpainting. The outer inpainting block maintains the original K-SVD algorithm, and is used to find a good dictionary fit for existing pixels. This step is optimum in the l2 norm error sense, i.e. it minimizes the estimation error of the existing pixels to find a dictionary expressing those existing image pixels. This dictionary, however, is not optimized for missing pixels. In fact, the missing pixels are not even considered in the optimization of the outer inpainting phase. Hence, another inpainting stage, called inner inpainting, is cascaded to the outer inpainting block to iteratively reduce the estimation error caused by the missing pixels. Since the dictionary generated by the outer inpainting block is optimal for existing pixels, it is also used as an initial dictionary for the inner inpainting block. The process continues unless the estimated maximum error is increased. The iterative modifications, though, are only applied to those missing pixels rather than to the whole image. The new algorithm shows significant improvements with respect to K-SVD both in terms of PSNR and visually.

Published in:

Multimedia Signal Processing, 2009. MMSP '09. IEEE International Workshop on

Date of Conference:

5-7 Oct. 2009