Automatic image registration and defect identification of a class of structural artifacts in printed documents

The work in this thesis proposes a defect analysis system, which automatically aligns a digitized copy of a printed output to a reference electronic original and highlights image defects. We focus on a class of image defects or artifacts caused by shortfalls in the mechanical or electro-photographic processes that include spots, deletions and debris missing deletions. The algorithm begins with image registration performed using a logpolar transformation and mutual information techniques. A confidence map is then calculated by comparing the contrast and entropy in the neighborhood of each pixel in both the printed document and corresponding electronic original. This results in a qualitative difference map of the two images highlighting the detected defects. The algorithm was demonstrated successfully on a collection of 99 printed images based on 11 original electronic images and test patterns printed on 9 different faulty printers provided by Xerox Corporation. The proposed algorithm is effective in aligning digitized printed output irrespective of translation, rotation and scale variations, and identifying defects in color inconsistent hardcopies