All the features used for veri"cation had almost the
same degree of e!ectiveness whether they were used to
compare the sample to prototype or vice versa.
f Central line features contain the largest number of
points per image. Some of these points are useless
because they are concentrated in the central region.
f Corner curve features are not highly e!ective in the
veri"cation process.
f It is quite su$cient to use the corner line features,
central circle features and critical point features in the
discriminating feature set.
f An average of 98% overall veri"cation certainty was
achieved.
f The best-"t method produced better results than existing
methods because each point was given a grade
according to its degree of matching.
f The decision to accept or reject was controlled by the
grade of the signature.
f The most powerful features were the critical point
features, although they are computationally exhaustive.
f The thinning algorithm may reduce the reliability of
the system because all the features are based on t