Graphics Reference
In-Depth Information
Tabl e 5. 6 Average bit rate savings of HEVC interpolation filter for the
luma component using HM6.0
BD-rate [%]
Class
Sequence
RA-Main
LB-Main
LD-Main
A
Traffic
2:4
n/a
n/a
(2560 1600)
PeopleOnStreet
0.1
n/a
n/a
Nebuta
0.4
n/a
n/a
SteamLocomotive
1:0
n/a
n/a
B
Kimono
1:8
2:6
0.6
(1920
1080)
ParkScene
2:7
4:5
2:2
Cactus
1:3
2:9
0:8
Basketball Drive
2:0
3:1
0:2
BQTerrace
5:0
7:1
0.5
C
Basketball Drill
2:5
3:6
2:1
(832
480)
all
4:3
5:5
2:7
PartyScene
10:7
11:9
10:5
RaceHorses
1:2
2:2
0:1
D
Basketball Pass
1:8
2:7
1:0
(416
240)
BQSquare
21:6
21:7
18:0
BlowingBubbles
7:5
9:1
7:6
RaceHorses
1:8
3:4
2:3
E
FourPeople
n/a
2:0
0.6
(1280
720)
Johnny
n/a
6:8
2:2
KristenAndSara
n/a
3:9
0:5
Average
4:0
4:9
2:6
and the results are compared. This experiment is conducted to see how much gain
collectively all the improvements the HEVC interpolation filter brings. The test
conditions can be summarized as:
￿
Four quantization values used: 22, 27, 32 and 37
￿
A total number of 24 different sequences are coded. These sequences are divided
into different classes that represent different use-cases and video characteristics.
￿
Tests are conducted for three different prediction structures: Random Access,
Low Delay with B pictures and Low Delay with P pictures.
￿
The coding efficiency is measured by using the Bjøntegaard-Delta bit rate
measure [ 2 ].
The detailed results are shown in Table 5.6 for the luma component, where it
is shown that on average, the interpolation filter of HEVC brings 4.0 % coding
efficiency gain. The results for the chroma component is also summarized in
Tab le 5.7 where the average coding efficiency gains reach 11.27 %. For some
sequences, especially for those that contain more high frequency content, gains
 
Search WWH ::




Custom Search