3D Holoscopic Video Representation and Coding Technology - Novel 3D Media Technologies - page 89

5.5.1 Experimental Results for Different 3D Holoscopic
Representation Formats

Although the self-similarity compensated prediction is able to explore the intrinsic
correlation between neighboring micro-images to improve the coding performance,
it is also important to understand how the coding efficiency depends on the
representation format of the 3D holoscopic content. Hence, the representation
that presents better RD performance would be preferred.
As described in Sect. 5.3 , besides the representation based on micro-images, it is
also possible to arrange the 3D holoscopic content based on an array of viewpoint
images and ray-space images. Therefore, a comparative analysis of the self-
similarity compensated prediction method applied to different forms of represen-
tation of the 3D holoscopic content is presented. Moreover, solutions proposed in
[ 10 ], where PVSs of micro-images, viewpoint images, and ray-space images are
encoded, are also compared. However, instead of encoding with H.264/AVC (as in
[ 10 ]), the HEVC codec is used, so as to have a fairer comparison with the other
evaluated solutions.
In summary, seven solutions are tested and compared:

1. HEVC : The original 3D holoscopic image is encoded using HEVC with “Intra,
main” configuration [ 29 ].
2. 3DHolo : The original 3D holoscopic image is encoded using the self-similarity
compensated prediction with “Intra, main” configuration [ 29 ]. A search range of
128 is allowed in this case.
3. VI-based 3DHolo : From the original 3D holoscopic image a VI-based
holoscopic image is generated, which is then coded using the 3DHolo coding
scheme with “Intra, main” configuration [ 29 ]. In this case, a larger search range
(312, for Plane and Toy , and 164, for Laura ) is allowed, since the samples with
significant correlation appear in a periodic structure given by the size of the
micro-lens array (see Sect. 5.3.2 ). Therefore, a proportional search range is used,
based on sizes of micro-lens array and micro-images. This guarantees a fairer
comparison between these different representations.
4. RI-based 3DHolo : The original 3D holoscopic image is pre-processed so as to
generate the corresponding RI-based holoscopic image, which is then coded
with the presented 3DHolo scheme. In this case, a search range of 312, for Plane
and Toy , and 164, for Laura , is used.
5. MI-based PVS : This solution corresponds to the solution proposed in [ 10 ] and
referred to as MI-based PVS. In this case, a PVS of micro-images is inter coded
using HEVC. Since the resolution of each micro-image is usually considerably
small, the largest coding block was set to 16 16 to avoid extra signaling

[ 18 ]. The MI-based PVS is then encoded using the “Low Delay, main, P slices
only” configuration [ 29 ]. Various orders for scanning the array of micro-images
were tested (raster, parallel, zigzag, and spiral), but only the spiral order is
presented as it achieved the best performance.