Image inpainting with nonsubsampled contourlet transform

A new algorithm for image inpainting based on the searching of redundancy for corner points across different scales and orientations is proposed. The searching utilises the nonsubsampled contourlet transform (NSCT) of the original image. The target region is filled-in following the priority which is given to the corner points that accumulate a high energy at different orientations and scales, and also having a high level of confidence. In each iteration, the source patch chosen is copied onto the target patch producing the minimum dispersion across different scales and orientations of the NSCT within it. This approach - to firstly fill-in corner points of high energy and confidence which minimize the dispersion in the target patch across different scales and orientations - have been tested on two groups of images, one group with broken edges and objects and a second group with large objects to be erased. The proposed algorithm was compared with three other methods and results obtained indicate that the capability of this algorithm is sustainable. (c) 2013 Elsevier B.V. All rights reserved.