Contour-based correspondence using Fourier descriptors

A contour correspondence problem, in which a contour is described in terms of epipolar lines of which the slope and the intercept functions are expressed as Fourier series, is addressed. The Fourier coefficients are computed via an iterative algorithm for which a condition is derived to test convergence. The contour correspondence between two images is solved based on the authors' proposition for the minimal spectrum distance (MSD). In the MSD scheme, either manual selection of the weighting factor can be adopted or a 'fuzzy and' operation-based automatic enhancement procedure can be employed for contour matching. Experiments are conducted for a set of scattered objects, a plant, a building, several Chinese characters and a group of line segments. The results show that for all of these contours, the iterative algorithm converges very quickly and the correspondence of the contours is accurate.