Digital Signal Processing Reference
In-Depth Information
With the proposed HVQ framework, we can reconstruct the image hierarchically
by the different codebooks which are generated by different layers of the image.
4
Experiments
In this section, we illustrate the performance of our hierarchical clustering VQ
framework. Firstly we give an example of the codebook generation, and then we use
our framework on some basic testing images and compare our framework result with
that of traditional algorithms.
For the codebook generation by executing the FCM algorithm, we can generate
different codebooks varying from codeword dimensions and numbers. The different
codebook size can be chosen as
M
=
16, 32, 64, 128, etc
and the different dimen-
sions of codewords can be
D
=×
2
2, 4
×
4, etc
. Fig. 2 demonstrates an example of a
codebook whose size is
M
=
32
and dimension is
.
D
=×
44
Fig. 2.
A codebook with size
M
=
32
and dimension
44
×
Then we take 256
×
256
grayscale Lena image as an example to illustrate our hie-
rarchical clustering VQ framework. By using the codebooks generated by the FCM
algorithm, the experimental PSNR results of the reconstructed image are given
in Table 1, with three codeword sizes
(2
×××
2, 4
4, 8
8)
and varying vector
dimensions (from 16 to 256).
Table 1.
The experimental PSNR results of the reconstructed image with three codeword sizes
(2
××× and varying vector dimensions (from 16 to 256)
2, 4
4, 8
8)
codeword dimension
codebook size
2
×
2
4
×
4
8
×
8
16
28.74
24.78
20.91
32
30.37
25.41
21.27
64
31.93
26.05
21.96
128
33.47
26.43
22.42
256
35.00
26.80
22.72
