Boundary normalization for recognition of non-touching non-degraded characters

Abstract

Presents a new approach to the recognition of non-touching, non-degraded characters. Our methods works on boundary functions of characters and attempts to normalize these functions to orientation and size variations of the symbols. This allows us to apply any 1D transform in order to extract invariant features. We validated this approach by performing several experiments on both printed and handwritten characters. The tests show that our method compares favorably to other methods described in the literature, both from the recognition accuracy and processing speed points of view.