Registration techniques for image subtraction

摘要

Image subtraction can be used to aid in detection of flaws in a manufactured product by subtracting the image of a perfect product from that of the sample under tests. For this technique to be successful, the two images must be precisely registered. Several registration techniques were applied to various images which had been subjected to translation, rotation, and distortion. It was found that a measure of the amount of translational misregistration could be determined to subpixel accuracy by measurement of binary or gray-scale centroids. Of three techniques for determining rotational misregistration, the measurement of the angle of the binary major axis was the most reliable.