Image Processing Reference
In-Depth Information
Fig. 3.13 The relation between the current macroblock and the epipolar lines in the reference
views
(
a, b, c
)
T
where F is the (3
3) fundamental matrix and l
¼
represents the line
having the form
ax + by + c
0. F can be found using the camera projection
matrices or the methods described in [
48
]. This relation suggests that the disparity
estimation area can be further reduced to a small region around this line. Figure
3.13
shows an example of the current view (view 1) and its references (view 0 and view
2) of the
Ballet
sequence. Results given in [
49
] show that a speed up of 21.8 times
can be achieved if the epipolar line is considered together with the disparity vector
discussed above.
The geometry methods were further enhanced in [
50
] by using an adaptive search
area along the epipolar lines based on the largest depth variation within the
macroblock being encoded. A speedup of 32 times is reported over the exhaustive
full search algorithm. The equation used to determine the search area is given by:
¼
SA
¼
min 3
þ ʱ
Δ
depth
10
where
Δ
depth
is the absolute difference between the statistical maximum and
minimum depth value within the sub-macroblock, while
is sequence dependent
and depends on the statistical variation of the depth within the sub-macroblock.
Using 8
8 pixel elements, this parameter is found as:
ʱ
ʱ ¼
=
˃
7
where
is the standard deviation of the differences between the statistical maxi-
mum and minimum depth values.
The authors in [
51
] propose an inter-component tool for the HEVC standard to
perform joint coding of the quadtrees in texture and depth videos. This technique
saves coding time and provides better compression, hence reducing the data that
needs to be transmitted. When the depth video is encoded before the texture video,
the quadtree structure for the texture is inherited from the already coded depth
quadtree. On the other hand, if the texture is encoded first, the depth quadtree
inherits the information from the texture.
˃
