Information Technology Reference
In-Depth Information
The composition of MPEG-2 video coding is shown in Fig. 30.
S
S
2
C
(
u
)
C
(
v
)
N
x
1
0
N
y
1
0
ª
(
x
1
u
º
ª
(
y
1
v
º
¦¦
F
(
u
,
v
)
f
(
x
,
y
)
cos
cos
¬
¼
¬
¼
N
2
N
2
N
8
㬍
8 block DCT
Adaptive
scan
DCT/
Quant.
Inv.Quant./
IDCT
Frame/Field DCT
Quantized
Coeff.
Entropy
coding
㪇
Motion
compensation
Reference picture
Motion
information
Intra / Inter
(
㶎
)
24
8
2
1
0
0
0
0
6
0
0
0
0
0
0
0
4
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
Ყセ丵
ㆬᛯ
Comparison
selection
•
Zero-run length
Non-zero value
Motion
Pred.
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
MIX
0
0
0
0
0
0
0
0
Half-pel MC
2D
VLC+EOB
Frame/Field adaptive pred.
(
㶎
) “Foreman” (MPEG standard image)
Fig. 30
MPEG-2 Video Coding
3.3.2 MPEG-4 visual
MPEG-4 visual is an international standard for the purpose of the coding at low bit
rate for mobile equipments. It is used in mobile devices such as cellular phones
and portable video players. MPEG-4 mainly performs coding and transmission for
progressive video signal from QCIF (176 pixels x 144 lines) to VGA (640 pixels x
480 lines) at lower bit rates of about 1 - 3 Mbit/s. The main features of MPEG-4
visual are shown below.
First, Intra Frame Prediction is performed for DC and AC data of transformed
and quantized coefficients. The entropy of a symbol which should be coded as
DCT coefficients can be reduced by prediction since the DC coefficient is equiva-
lent to the average value in a block and AC coefficients including low frequency
harmonics have high spatial correlations. The outline of Intra Frame Prediction in
MPEG-4 is shown in Fig. 31.
Next, in addition to half-pel accuracy of Motion Compensated Prediction,
MPEG-4 also supports quarter-pel accuracy prediction which uses the virtual
samples between half-pel pixels as a candidate of the prediction. Additionally, a
16x16 pixel macroblock domain can be equally divided into four 8x8 sub-blocks
and Motion Compensated Prediction can be adaptively performed in a 16x16 mac-
roblock unit or the unit of an 8x8 sub-block. With this technique, the performance
of prediction for complicated motions within a macroblock can be improved. Fur-
thermore, three dimensional VLC is performed on the quantization indexes after
