Information Technology Reference
In-Depth Information
S
(
i
,
j
)
F
(
u, v
)
(Low)
(Low)
Horizontal
spatial
frequency
Vertical
spatial
frequenc
Horizontal
Vertical
(High)
(High)
DCT
Image signal tends to
have high-power in
low-frequency
components.
IDCT
i
j
u
v
8 pixels
8 pixels
㸣
DCT
DCT can effectively
concentrate essential
information to be
coded.
IDCT
“Lenna”
㨇
Pixel domain
㨉
㨇
Transform domain
㨉
Fig. 14
Example of transformation by DCT
Avoid visual degradation
by means of fine quantization
in low frequency
F(u, v)
.QY
.QY
24
8
2
1
0
0
0
0
*QTK\QPVCNURCVKCN
HTGSWGPE[
8GTVKECNURCVKCN
HTGSWGPE[
6
0
0
0
0
0
0
0
*KIJ
4
0
0
0
0
0
0
0
*KIJ
1
0
0
0
0
0
0
0
Not
coded
zeros
in high
frequency
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
u
v
0
0
0
0
0
0
0
0
Encode the combination of “number of sequential zeros” and “non-zero
coefficient” obtained by zigzag scan from low to high frequencies
䋨0,24䋩,䋨0,8䋩,䋨0,6䋩,䋨0,4䋩,䋨1,2䋩,䋨0,1䋩,䋨2,1䋩,EOB䋨End Of Block䋩
Use Huffman codes corresponding to occurrence probability of events
Fig. 15
Example of Zigzag Scan and Run Length Coding
3.2.3 Quantization
Quantization is a technique of reducing the amount of information directly, and there
are mainly two methods well-known for video compression, which are Scalar Quanti-
zation and Vector Quantization. Scalar Quantization is an operation of making an in-
put signal correspond to one of k kinds of values which are represented as
q
1
, ...,
q
k
as
shown in Fig. 17.
