A Graph Matching Approach to Symmetry Detection and Analysis - Computational Intelligence Paradigms in Advanced Pattern Classification - page 133

(a) (b)

(c) (d)

Fig. 10 Rotational near-symmetry with clutter and low repeatability of interest regions. ( a )
Interest points. ( b − d ) Detected rotations.

(a) (b)

Fig. 11 ( a ) Failed symmetry detection. The corresponding points are not symmetrical in the
real world, but only within the image. ( b ) Correct symmetry detection

applications, one can add task-specific geometrical constraints and improve the re-
sults. Figure 12i shows another failure of the SSA, where a false symmetry axis is
detected. Note that a valid axis was detected in the same image in Fig. 12f. The
failure is due to the detection of spurious feature points, on the car's hood.