We describe an algorithm and its implementation for improving the position detection accuracy of a lightpen. The algorithm consists of computing, for every field of video, the centroid of the patch of light seen by the lightpen, and using it to determine the location of the lightpen on the screen. The centroid is found by computing the zero- and first-order moments of the patch of light. Compared to simple thresholding, which is used in most lightpens, this results in improvement of the accuracy of the location by at least a factor of 40. It is possible to resolve better than one quarter pixel distance on a screen of 1000 multiplied by 1000 pixels. The pen is thus capable of providing inputs for drawing antialiased lines by humans.
