Invariant Shape Descriptors for Handwritten Strokes

Abstract

In this work, we present a novel approach to extract rotation, translation and scale invariant shape descriptors for handwritten strokes. Invariant descriptions of shapes are useful for shape matching and retrieval, where shapes can be randomly rotated, translated and/or scaled. Using a minimal set of segment types and novel corner-point descriptors, we describe an algorithm to extract invariant descriptions for any planar outlines. The suitability of the proposed descriptors in shape matching is shown through a number of experiments.