Partial Shape Recognition using Fourier-Mellin Transform Methods

Abstract

In recognizing shapes automatically by computer, a problem often arises when the unknown object is partially obscured or poorly segmented. Most algorithms described in the literature use syntactic methods. A major problem is describing a proper grammar for several objects. This problem is even further complicated when each object may be imaged from many possible aspect angles. The algorithm in this paper avoids this problem. It uses a global method to normalize scale and starting point, even when part of the contour is missing or incorrect.