BORDER FOLLOWING - NEW DEFINITION GIVES IMPROVED BORDERS

Border following is widely used in the preprocessing of many binary images. Images may be taken to consist of a set of black objects on a white background, or vice versa, and the objects may have holes in them; some of the holes may contain objects, and this may be repeated. Finding the borders of the objects allows considerable compression and has other advantages, but is more difficult than may appear at first sight. In particular, it is not difficult to obtain algorithms which produce re-entrant curves as candidate borders, and others which produce borders which are unsatisfactory for various reasons. The paper describes a co-recursive algorithm obtained from a new definition of borders. Because of some counter-intuitive aspects of the subject, it was necessary to prove that the algorithm produces a border in the sense of the paper. Experiments on a variety of images are described, and the results show that the borders described in the paper are generally smaller and better connected than some others.