3D Video Representation and Coding - Novel 3D Media Technologies - page 36

chain code, and simplified depth coding [ 23 , 32 - 34 ]. Depth Lookup Table [ 33 , 35 ]
reduces the bit depth of the each depth-map sample, achieving good compression
performances for highly quantized depth-maps, which are commonly used. View
Synthesis Optimization (VSO) [ 33 , 34 ] performs rate-distortion optimization for
the compression of the depth-map samples based on the distortion between two
intermediate views, synthesized with the original and the compressed depth-maps.
The use of intra prediction is a powerful tool for image compression and has also
been adopted for DC. Due to the differences between texture and depth-map
images, several proposals have been made for intra prediction modes specifically
tailored for DC. These modes can be used to improve the performance of traditional
intra prediction framework of H.264/AVC or H.265/HEVC. Intra prediction modes
for DC are designed to efficiently represent both the homogeneous areas as well as
the edges, which are commonly found in depth-map macroblocks (MB).
3D-HEVC uses the previously mentioned depth modeling modes (DMM) [ 23 ,
34 ], together with the original HEVC intra prediction modes. Furthermore,
3D-HEVC also defines an alternative way to approximate the prediction residues,
referred to as constant partition value (CPV). CPV can be used as an alternative to
the traditional transform coding. DMM assume that the depth-map blocks are
composed by two uniform regions, divided by one or more edges. Explicit infor-
mation is transmitted to signal the MB structure, as well as the values of the uniform
areas. Two types of partitions are defined, namely Contours and Wedgelets.
Wedgelets (see Fig. 3.11 ) assume that a single linear edge divides the two uniform
areas, P 1 and P 2 . 3D-HEVC defines two wedgelet modes: one explicit and one
implicit mode. The former transmits the information about the edge location and
values of the uniform regions, while the latter predicts the wedgelet partition based
on the texture information. The contours-based prediction modes (see Fig. 3.11 ) are
able to represent more complex arrangements of arbitrarily-shaped regions. The
complex region partition is determined based on the texture block in order to avoid
the transition of this information. The values of the uniform depth regions, P 1 and
P 2 , are predicted from previously reconstructed blocks and represented using a
Depth Lookup Table (DLT) [ 33 , 36 ].
Fig. 3.11 The wedgelet ( left ) and contour ( right ) depth modeling modes of 3D-HEVC [ 12 ]