Graphics Reference
In-Depth Information
Fig. 5.6
Detail of a video sequence exemplifying the merge concept. (
a
) In the foreground, the
scene contains a moving object (a pendulum) with motion indicated by the
arrow
.(
b
)Arate-
distortion optimized quadtree partitioning for inter-picture prediction parameters is indicated by
the
white borders
.(
c
) Only effective block borders are shown, i.e. borders that separate blocks
with different motion parameters. Reproduced with permission from [
12
], © 2012 IEEE
5.2.2
Inter-picture Prediction Block Merging
In image or video compression it is very reasonable to deploy a block based
image partitioning mechanism in order to apply different prediction models to
different regions of an image. This is because a single model can in general not
be expected to capture the versatile characteristics of a whole image or video.
HEVC uses a quadtree structure to describe the partitioning of a region into sub-
blocks. In terms of bit rate, this is a very low-cost structure while at the same
time, it allows for partitioning into a wide range of differently sized sub-blocks.
While this simplicity is an advantage for example for encoder design, it also bears
the disadvantage of over-segmenting the image, potentially leading to redundant
signaling and ineffective borders. This drawback is effectively addressed by block
merging as explained in Sect.
5.2.2.1
. A description of the exact algorithm and
bitstream syntax follows in Sects.
5.2.2.2
-
5.2.2.5
.
5.2.2.1
Background
An example partitioning using the quadtree structure is shown in Fig.
5.6
afora
uni-predictive slice in an HEVC-encoded video. As can be seen, the area around
the pendulum is heavily partitioned in order to capture the motion in front of the
still background. Figure
5.6
c shows the same partitioning, but without ineffective
borders, i.e. borders dividing regions of equal motion parameters. It becomes
evident that in this particular situation, the quadtree structure is unable to accurately
capture the motion without introducing ineffective borders. It is easy to see that
this over-segmentation easily occurs whenever moving objects in a scene cause
