An automatic visual inspection method of pattern using a circular feature extraction filter

The purpose of this paper is to describe a method for automatic visual inspection of patterns, such as bare-printed circuit boards (PCB's). The method detects faults of patterns by comparing an input pattern to be inspected with the reference one after features extraction from each of them. For the feature extraction, a circular local filter in the method is used to recognize the pattern's interior points, exterior points, boundary points, the boundary's direction, and the boundary's curvature. The ability of the filter that detects the difference between two boundaries of different curvature is also presented.