Full Parallax 3D Video Content Compression - Novel 3D Media Technologies - page 65

Fig. 4.17 Inter-view
derivation of a second DV

Table 4.3 BD-rate variations
for improved NBDV and
IVMP using one DV for each
inter-view reference picture

Reference: 3D-HEVC (HTM7.0 without modifications)
3 3 views 11 5 views

Coast Akko Coast Akko

Adaptation only 1.1% 2.3% 2.4% 3.3%

1.2% 2.4% 2.7% 3.7%

BiDMV

2 IPMC 1.1% 2.3% 2.8% 3.5%

Both 1.3% 2.5% 3.1% 3.9%

Table 4.4 BD-rate variations
for improved NBDV and
IVMP with inter-view
derivation of a second DV

Reference: 3D-HEVC (HTM7.0 without modifications)
3 3 views 11 5 views

Coast Akko Coast Akko

BiDMV + derivation 1.9% 2.9% 3.4% 3.9%

2 IPMC + derivation 1.3% 2.4% 2.8% 3.5%

2.0% 2.9% 3.9% 4.2%

Both + derivation

software version 7.0 (HTM7.0). The test conditions are the same as in Sect. 4.3.2.2 ,
(i.e., allowing two dimensional structures configuration). The Central2D structure
(proposed in Sect. 4.3.2 ) is used in all the following experiments. The reference is
HTM7.0 without software modifications.
In Table 4.3 , the BD-rate gains provided by the adaptation of NBDV and IVMP
to a two dimensional structure are up to 3.3%. It confirms the impact of the use of
horizontal and vertical dimensions at the inter-view references structure level. The
insertion of a second IPMC in the merge candidate list and the bi-prediction for the
DMV merge candidate separately increase the gains up to 2.4% for the 3 3 views

configuration and 3.7% for the 11 5 views configuration. The combination of

both provides a gain up to 2.5% and 3.9%, respectively, with 3

5 views. The combination of both tools provides slightly higher gain than the sum
of each taken separately because the bi-prediction allows NBDV to find more often
a second DV, which increases the chances to create a relevant second IPMC
candidate. Table 4.4 shows that the proposed derivation for the second DV is

3 and with 11