Low-level and high-level correlation for image registration (IC inspection)

A method based on the modification of the two-level correlation method and the preliminary results of its application to an IC inspection problem are presented. To reduce the computational complexity, a fast correlation method is that uses random shifts, random pixel selection within the search image, and optimal binarization of gray levels. This generates the possible locations where the template has high correlation peaks within the search image. As an example, the four corners of the IC can be located using the correlation peaks. Because of the random search procedure, there is uncertainty involved in matching the location correlation peaks to the corners of the IC. An evidence scheme based on the Dempster-Shafer formalism is used in the second level (high-level correlation) to resolve the uncertainty. >