A method for invariant pattern recognition using the scale-vector representation of planar curves

Abstract

This article presents a new process for shift, rotation, and scale invariant pattern recognition using shapes. We represent the shape by using its two-dimensional contour, and we describe this planar curve not at all different scales, but each one of parts in the curve isolating a different structure at a single scale. A scale-vector representation of contours usually avoids missing fine features and overlooking coarse features. A polygonal approximation of these planar curves can be made by joining the successive dominant points detected on the contour represented at its scale vector. Dominant points of digitized curves are points with high curvature value. We present a scale-vector-based dominant point detection algorithm which needs no input parameter and remains reliable even when features of multiple size are present on the digital contours. Model-based recognition is achieved by comparing the polygonal approximation to the contour extracted from shape A, which is stored as a model for some particular object, with the polygonal approximation to the contour extracted from shape B, which is found to exist in an image. To compare polygons we use the L 2 distance between the turning functions of the two polygons. This method to compare polygons is invariant under translation, rotation and change of scale, taking time O( mnlog mn) to compare an m vertex polygon against an n vertex polygon.