A Novel Exemplar based Picture Completion Scheme using Adaptive Patch Size
To maintain our historical monuments, to regenerate any distorted image data etc., it is essentially desired to
complete the damages region of image with non-detectable visual perception. This paper introduces a novel approach of
picture recreation using non-local mean and adaptive patch size. Earlier, recreation approaches have considered only the
surrounding elements of the target region, to fill up the lost region which was unsatisfactory in most cases. These approaches
uses fixed patch size which does not take into account the local picture characteristics of the target patch. The proposed
approach is basically an exemplar based picture recreation technique that uses non-local data in a very constrained way. In
this algorithm, an exemplar based adaptive window, which is adjusted according to the picture itself, is used to fill the
damaged region. Moreover, during target patch replenishment, the database is used from multiple specimens within the
picture. An alteration of a target pixel is determined using weight similarity technique and assembles to form a lost database.
Experiments have been performed which shows that the proposed technique produces better quantitative and qualitative
recovery as comparing to a present exemplar basis technique.
Keywords - Picture creation, Nonlocal-means, Patch size, Image in-painting.