Rendering, Adaptation and 3D Displays - 3DTV: Processing and Transmission of 3D Video Signals - page 134

Left reference view

Interpolated view

Right reference view

Foreground
Background
Pixels interpolated bidirectional - wighting coefficients used for the reference views
Pixels occluded in one of the reference views - unidirectional interpolation
Pixels interpolated bidirectional - pixels visible in both reference views

Figure 5.8 Interpolation scheme taking into consideration the occlusions

with unidirectional interpolation and the pixels filled with bidirectional
interpolation. The disparity map may not perfectly match the borders of
video objects.
Most of the time, it is difficult to track the transition from a background
region to a foreground region in a sharp way due to the slightly blurred
edges, which therefore leads to kind of a disturbing ghosting effect in
the interpolated view images unless some caution is taken. In one of the
recent European projects, called ROMEO (Remote Collaborative Real-Time
Multimedia Experience over Future Internet) [4], the transitions between
the unidirectional and the bidirectional interpolation are performed in a
weighted manner depending on the pixels' location as a cautionary step
during the interpolation process. In other words, the coefficients used
for bidirectional interpolation are adapted based on the position of the
occluded pixels. The coefficient management depends both on the width
of the occlusion space and the depth difference between the background
and the foreground layers. Figure 5.8 illustrates the interpolation process
taking into account occlusion handling and bidirectional interpolation with
weighting factors.
Another visual degradation that results after the view interpolation is the
aliasing effect, which is visible on the borders of the synthesized objects.
The reason for this is the association of a single disparity value to each
of the interpolated pixels. Thus, the borders of the synthesized objects in
the interpolated disparity map sharply coincide with the pixel borders and
so do the object borders in the texture of the interpolated view. Aliasing
should be efficiently overcome by means of post-processing. In the first
step, the borders need to be detected in the interpolated disparity map and
subsequently a sub-pixel mask needs to be adapted, based on the content
in order to filter the contours. As a result, each sequence of pixels along
the object borders is associated with a couple of disparity values, like semi-
transparent objects. Figure 5.9 shows an example of a rendered texture,
before and after de-aliasing.