The average ridge distance of fingerprint images is used in many problems a
nd applications. It is used in fingerprint filter design or in identificati
on and classification procedures. This paper addresses the problem of local
average ridge distance computation. This computation is based on a two-ste
p procedure: first, the average distance is defined in each significant por
tion of the image and then this information is propagated onto the remainin
g regions to complete the computation. Two methods are considered in the fi
rst step: geometric and spectral. In the geometric approach the central poi
nts of ridges are estimated on a regular grid and straight lines passing th
rough these points and parallel to the ridge directions are used. The secon
d method is based on the computation of harmonic coefficients leading to ef
fective estimates of the average ridge period. In order to complete the ave
rage distance map a diffusion equation is used so that maps with minimum va
riations are favored. Finally, some experimental results on NIST SDB4 are r
eported. (C) 1999 Patern Recognition Society. Published by Elsevier Science
Ltd. All rights reserved.