Page 244 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 9.7. Postprocessing (or Concealment) Techniques 221

coe$cients of the four or eight neighboring blocks. Another approach is to
form a partial DC value at each boundary by taking the average of a one-,
two-, or four-pels-wide neighborhood. The recovered DC coe$cient is then
the average or the median of the four partial DC values.
In Ref. 193 the lost DCT coe$cients of an intracoded block are recovered
by minimizing the intersample variation within the block and across the block
boundaries. This is based on the smoothness property of image and video
sequences. In Ref. 189 the same method is extended by adding a temporal
smoothness measure.
Another property that is used in error concealment is edge continuity. Thus,
if the direction of an edge in a neighboring block indicates that the edge
passes through the damaged block, then the concealment process must con-
serve the continuity of this edge. For example, in Ref. 194 an edge classi er is
applied to the neighboring blocks to determine which directions characterize
the strongest edges passing through the damaged block. For each of these clas-
si ed directions, directional spatial interpolation along the respective direction
is used to create a block from the neighboring pels. The blocks are then
mixed together in such a way that all the strong edge features are preserved
and combined in a single block used for concealment.
Statistical correlation is another a priori assumption utilized in error
concealment. For example, in Ref. 195 the pel values of a frame are
modeled as a Markov random eld (MRF). Maximum a posteriori proba-
bility (MAP) estimation is then used to spatially interpolate the damaged
blocks.

9.7.2 Temporal Error Concealment
Temporal techniques exploit the high temporal correlation of video signals and
conceal damaged pels in a frame using information from correctly received
and=or previously concealed pels within a reference frame. Such techniques
apply primarily to intercoded blocks. They may work for some intracoded
blocks but will completely fail in cases like scene changes and uncovered
background.
As in motion-compensated prediction, the process of temporal concealment
involves two stages: concealment displacement estimation and displacement
compensation, as shown in Figure 9.8(a). For this reason, temporal conceal-
ment is sometimes referred to as motion-compensated concealment.
Conventional temporal techniques estimate one concealment displacement
for the whole damaged block and then use translational displacement com-
pensation to conceal the block, as shown in Figure 9.8(b). Such techniques
perform very well when the original motion vector of the damaged block is
available. In this case the rst stage of the temporal concealment process,