Title: Pixel based Texture in Painting using Optimal Neighborhood Positioning
Publisher: Guru Nanak Publications
Series: Volume 3 Issue 1
Authors: Ravneet Kaur, Dan Hauer, Scott Umbaugh, and Robert Leander
Dermoscopy is an imaging technique used to
facilitate accurate automatic identification and
diagnosis of skin cancer. However, because of the
presence of noise in an image in the form of hairs and
bubbles, identification may lead to inaccurate results.
An in painting algorithm was developed for
reconstructing image backgrounds after the removal of
hairs and bubbles. The algorithm assumes a Markov
Random Field model for the surrounding texture and
uses pixel-based texture synthesis techniques to
determine the best value for an unknown target pixel,
based on a neighborhood search model. The algorithm
attempts to optimize the quality of in painting by
choosing an optimal neighborhood location around the
target pixel. Using mean–square error (MSE) metric,
three neighborhood (window) sizes were objectively
investigated for overall effectiveness of the algorithm.
Larger window sizes (e.g., 5, and 7) provided output
results with less visible artifacts. Thus, large window
size provides better background reconstruction.
Algorithm, Marko random field, Mean square error, Pattern based synthesis, Pixel, Synthesis, Texture.