Experiments with an automated visual inspection system

Experiments were performed with an automated visual inspection system that detects potential flaws by comparing reference and comparison images. Three main components evaluated are: 1) image-differencing-based ambient light compensation, 2) electronic-shuttering-based ambient light rejection, and 3) image registration. The results indicate that the ambient light compensation algorithm yields 0.01% to 0.4% false flaws with lab-simulated sunlight changing from 25% to 100% intensity. Electronic shuttering with synchronized strobe lighting reduces false flaws considerably. As exposure shortens from 1/60 (fully open) to 1/1000 second, false flaws decrease 100-fold from 0.4% to 0.004%. The current registration algorithm corrects a very limited range of misregistration, correcting approximately 4 pixels of pure translational shifts over the inspection surface. A more robust image registration algorithm that can correct both translational and rotational shifts over 10 pixels misregistration would be highly desirable.