
Figure 4. Comparison between No-updated and updated model.
blocks patch
i
, patch
j
is defined as:
We set an experienced threshold θ =0.6 for S(i, j). Calculate the sum N of the codebook elements
in the codebooks whose similarity is bigger than 0.6. If N >
m
2
, then put the new patch into the
codebook. To keep the balance of the number of the codebooks, we randomly discard a patch
in the codebook. Thus we get the new codebook and the leaf node’s p
L
and D
L
is computed
correspondingly. Finally, during this updating strategy, what can be assured is that the similarity
between the object model and the object status in the most recent period is the biggest.