Information Technology Reference
In-Depth Information
Vector Quantization is an operation which quantizes several samples at the
same time and expresses them with a representative vector which gives the best
approximation of the samples [16, 17]. The sources of information which consist
of many dimensions are quantized by one of the representative points of a multi-
dimension space by Vector Quantization. Therefore, Vector Quantization has the
following advantages,
(1)
Coding efficiency can be raised by adopting the correlation and the depend-
ency between the vectorized samples in the quantization mechanism.
(2)
Even if the vectorized samples are completely independent, the multi-
dimensional signal space can be divided into its quantized sections.
(3)
Samples can be coded with non-integer word size by assigning a quantized
represented vector or its codeword.
The key map and the principle of Vector Quantization are shown in Fig. 18 and
Fig. 19, respectively.
9JGPDNQEMKUSWCPVK\GF+PRWVXGEVQT
KUGZRTGUUGFYKVJ
x
=
{
x
1
,x
2
,…,x
16
}
x
2
x
1
x
2
x
3
x
4
x
4
x
5
x
6
x
7
x
8
RKZGNU
x
9
x
10
x
11
x
12
x
13
x
14
x
15
x
16
x
㧦+PRWVXGEVQT
x
k
RKZGNU
y
i
㧦1WVRWVXGEVQT
㧔4GRTGUGPVCVKXG3WCPVK\CVKQPXGEVQT㧕
x
1
x
3
Fig. 18
Key map of Vector Quantization
3.2.4 Entropy Coding
Entropy Coding is a method of describing the mode information, motion vector in-
formation, quantized values, etc. as a series of binary signals which consists of
only 0 and 1 (binarization). The total amount of codes is reducible by assigning
coded words according to the occurrence probability of symbols. Huffman coding
and arithmetic coding are typical entropy coding methods used in video coding.
Huffman coding is a method of designing and using a variable length code table
which associates symbols and code-words. This method can shorten the average
code length by assigning short codes to symbols with high occurrence probability
and long codes to symbols with low occurrence probability. An example of Huff-
man coding is shown in Fig. 20.
