Image Processing Reference
In-Depth Information
Fig. 1.11
The result of chaotic scan for
K
=
10000 and optimal radius
r
=
2
.
7
of the original message is done on a certain number
b
of bit-planes from the original
message (e.g. an image).
Within this section we consider the message signal to be identical to the original
one i.e.
K
=
N
. For each block, a search through the codebook reveals the best match
(associated to a label encoded with log
2
D
bits) and consequently in the decoding
stage (assuming the existence of the same codebook in the Rx unit) the initial block
is replaced with one of the
D
code-vectors in the codebook. Consequently, each bit-
plane may be treated independently i.e. algorithm parameters such as the CA rule
(ID) generating the codebook, its size
D
, the block size
m
, may be optimized such
that the bit error (
B
err
)
0
denotes the most significant plane, there is also an option in choosing the exact num-
ber
b
of bit-planes knowing that the most important in term of overall performance
(measured as the PSNR of the recovered image with respect to the original one) is
bit-plane 0. The compression performance in this case is always expressed in
bits
per pixel
(
bpp
). For instance, while using for each
b
for that specific bit-plane is minimized. In the following
b
=
=
6
bit-planes
m
=
64
(8
×
8
blocks),
D
=
64, the rate is computed with the following general formula:
bpp
=(
b
log
2
D
)
/
m
(1.4)
For the above particular values (quite usual with this compression scheme) a 0.56
bpp rate results. An exemplification of the encoding and decoding stages for the
CA-based dictionary VQ system is given in Figures 1.12 and 1.13.
1.3.2
Learning CA-Based Dictionaries and Performance
Evaluation of the CA-VQ System
As indicated above, the most important aspect in optimizing the CA-based com-
pression scheme is the choice of the CA structure and its rule, since it generates
the dictionary. In previous works [21] [11] a guided search through the space of
Search WWH ::
Custom Search