Regularization of shortest-paths and active contours has been consider
ed and attempted by a number of workers. However, it was not until the
development of the ''time-delayed dynamic programming'' algorithm of
Amini et al. (1990) in the active contours context that a method was f
ound which was able to apply a simple and intuitive smoothness constra
int with an efficient computational scheme. We show that, when applied
to the shortest-path problem, this technique gives rise to a simple a
nd efficient algorithm. However, we find that the method is not practi
cally useful in some situations because of discretization effects. A m
odification using pixel subdivision is proposed which to a large exten
t overcomes this problem. The modified method is illustrated using two
examples: fracture detection in borehole images and road detection in
satellite images. (C) 1997 Elsevier Science B.V.