A Fast Exemplar-Based Image Inpainting Method Using Bounding Based on Mean and Standard Deviation of Patch Pixels

Jungmin SO  Baeksop KIM  

Publication
IEICE TRANSACTIONS on Information and Systems   Vol.E98-D   No.8   pp.1553-1561
Publication Date: 2015/08/01
Online ISSN: 1745-1361
DOI: 10.1587/transinf.2014EDP7369
Type of Manuscript: PAPER
Category: Image Processing and Video Processing
Keyword: 
image inpainting,  exemplar-based,  bounding,  

Full Text: PDF>>
Buy this Article




Summary: 
This paper proposes an algorithm for exemplar-based image inpainting, which produces the same result as that of Criminisi's original scheme but at the cost of much smaller computation cost. The idea is to compute mean and standard deviation of every patch in the image, and use the values to decide whether to carry out pixel by pixel comparison or not when searching for the best matching patch. Due to the missing pixels in the target patch, the same pixels in the candidate patch should be omitted when computing the distance between patches. Thus, we first compute the range of mean and standard deviation of a candidate patch with missing pixels, using the average and standard deviation of the entire patch. Then we use the range to determine if the pixel comparison should be conducted. Measurements with well-known images in the inpainting literature show that the algorithm can save significant amount of computation cost, without risking degradation of image quality.