While matching between images using critical points
requires an exhaustive combinatorial search, the other
features are constrained by the route (line or curve)
number, where the search is performed route by route
(Fig. 8). However, critical point features produce the best
veri"cation results when used alone (Fig. 9). Corner line
features and central circle features have a reasonable
number of representative points. Since central line features
su!er from a high-intensity central region and corner
curve features have the lowest veri"cation results, the
appropriate discriminating feature set consists of a combination
of the corner line features, central circle features
and critical point features.