Abstract
A method of threshold setting has been developed based on a multiple regression analysis in an automated visual inspection. The threshold is assumed to be a linear combination of the local difference, local brightness, local contrast, etc. of the two gray-scale images to be compared. The coefficients of the linear combination are determined by the multiple regression analysis using a scattered diagram of the two gray-scale images. By this method good performance can be realized to justify the threshold setting according to the patterns to be inspected even when the inspected patterns are changed. This method is confirmed to be useful for a comparison inspection of complicated LSI multi-layer wafer patterns as a result of some experiments.
