Different window configurations and measures for detecting edges betwe
en regions of different mean in synthetic-aperture radar (SAR) images
are compared. Two criteria for optimisation are considered: (i) maximi
sing the total probability of detecting an edge within a window; and (
ii) maximising the accuracy with which the edge position can be determ
ined. Clearly, an ideal edge detector would combine both properties. M
aximum-likelihood solutions to the two criteria are discussed and it i
s shown how they determine the choice of both window configuration and
the measure adopted. The performance degradation encountered when the
tests are applied to the scenario for which they are not optimised is
compared and the test which best meets both original criteria is iden
tified. Combination of the good features of each to yield the best ove
rall performance in a two-stage edge-detection scheme is discussed.