Precise localization of geometrically known image edges in noisy environment

A method based on random samples least variance geometric form determination in precision measurement using computer vision is presented. The method is designed to extract the geometrically known image edges in a digital image with subpixel precision. The advantages of such a method are that the localization precision is unaffected by the noisy environment, and for many elementary geometrical forms it does not need traditional edge-following but only vertical and/or horizontal scannings of images. The precise localization and the elimination of the noise effect are achieved by iterative form determination from randomly chosen image points and the selection of points according to the measurement of the distance between image points and the reference primitive. The application of this technique to the measurement of ellipse and polygonal family forms is presented