Image Processing Reference
In-Depth Information
As a result, the chosen block becomes the candidate predictor (see Fig.
5.13
) and
the relative position between the two blocks is derived as a self-similarity vector
(similarly to a motion vector). In the self-similarity compensation block, the inverse
quantized and inversed transformed prediction residual is added to the predictor to
form the reconstructed data that is stored in the prediction memory to be available
for future predictions.
5.4.3 Syntax Adaptation in a HEVC Extension for 3D
Holoscopic Scalable Coding
Extending HEVC to support 3D holoscopic scalable coding basically means to
introduce the IL and SS references into the buffer of reference pictures of the HEVC
and to allow them to be used by the existing inter-frame prediction modes. Conse-
quently, no changes in the lower levels of the syntax and decoding process of
HEVC are needed and it only involves small changes to the high level syntax of
HEVC, notably, in terms of additional high level syntax elements necessary to
encode 3D holoscopic content.
The following two important pieces of information that are carried through a
Sequence Parameter Set (SPS) extension in order to be available at the decoder side
were identified:
•
Acquisition information
: As discussed in Sect.
5.4.1
, this information is com-
posed of a set of parameters, which are necessary to build the IL reference (e.g.,
resolution of micro-images, structure of the micro-lens array, size, and position
of the patches).
•
Holoscopic dependency information
: Since two new references are included into
the picture buffer, this information is necessary to indicate which holoscopic
references (among IL and SS references) are available for each of the two
reference picture lists (similarly to HEVC). This way, it is possible to distinguish
the new reference frames from each other and from temporal references.
The holoscopic dependency information is used to build the reference picture
lists for a 3D holoscopic picture being coded. In this case, each list is initialized as
for conventional 2D video and could include any available temporal reference
pictures (temporal prediction is not addressed in this chapter). Additionally, both
the IL and SS references are included in the lists, becoming available for prediction
of the current picture.
It is also important to notice that, in terms of lower syntax levels (e.g., Prediction
Unit syntax level of the HEVC [
18
]), the decoding modules do not need to be aware
of the reference picture type, and the distinction is only done in the higher level.
Consequently, the prediction of a 3D holoscopic picture becomes adaptive in the
encoding side, in the sense that the encoder can select the best reference picture in a
Rate Distortion Optimization (RDO) sense, resulting in only an index of the
