Figure 13:
Performance of the nearest neighbor approximator in the MEF space.
The performance is given as a function of the number of MEF's used.
We show some experimental results to indicate the performance of the nearest neighbor approximator in the MEF space. We computed MEF's using 660 training sequences. Fig. 14 shows top 10 MEF's.
The number of
MEF's was selected based on the variation ratio
, where
m out of n MEF vectors were used,
as defined in [15].
Table 2 shows the number of MEF's corresponding to the
variation ratio.
Table 2: The number of MEF's vs. the variation ratio
Fig. 13 shows the performance of the nearest neighbor approximator under the different variation ratio. The performance first improves when the ratio r increases. Then, at the point r=0.4, the performance saturates at the recognition rate 87.0%.