Databases Reference
In-Depth Information
T A B L E 10 . 1
Training set for designing
vector quantizer codebook.
Height
Weight
72
180
65
120
59
119
64
150
65
162
57
88
72
175
44
41
62
114
60
110
56
91
70
172
T A B L E 10 . 2
Initial set of output points for
codebook design.
Height
Weight
45
50
75
117
45
117
80
180
set have been assigned to the fourth output. The distortion for this assignment is 387.25. We
now find the new output points. There is only one vector in the first quantization region, so
the first output point is (44, 41). The average of the four vectors in the second quantization
region (rounded up) is the vector (58, 102), which is the new second output point. In a similar
manner, we can compute the third and fourth output points as (64, 117) and (69, 168). The
new output points and the corresponding quantization regions are shown in Figure
10.8
.From
Figure
10.8
, we can see that, while the training vectors that were initially part of the first
and fourth quantization regions are still in the same quantization regions, the training vectors
(59,115) and (60,120), which were in quantization region 2, are now in quantization region 3.
The distortion corresponding to this assignment of training vectors to quantization regions is 89,
considerably less than the original 387.25. Given the new assignments, we can obtain a new set
of output points. The first and fourth output points do not change because the training vectors in
the corresponding regions have not changed. However, the training vectors in regions 2 and 3
have changed. Recomputing the output points for these regions, we get (57, 90) and (62, 116).
The final form of the quantizer is shown in Figure
10.9
. The distortion corresponding to the
final assignments is 60.17.
