Page 270 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 10.5. Temporal Error Concealment for Multiple-Reference 247

concealment techniques have been proposed in the literature, and their per-
formances have been extensively studied within typical single-reference video
codecs operating over various error-prone channels. There is, however, a need
to characterize the performance of such techniques within multiple-reference
video codecs. This is the main aim of this section.
Temporal error concealment within a multiple-reference video codec can be
split into two problems: spatial-components (d x ;d y ) recovery and temporal-
component d t recovery. Thus, a multiple-reference temporal error concealment
method can be represented by a combination of the form S-T, where S is the
technique used to recover the spatial components and T is the technique used
to recover the temporal component. In this section, S and T can be chosen
from the following list of techniques

ZR The recovered motion component (either spatial or temporal) is set to
zero. In Chapter 9 this was referred to as temporal replacement (TR).
AV The recovered motion component is set to the average of the correspond-
ing components of a set of neighboring motion vectors. In this section,
four neighboring vectors are used: top, bottom, left, and right.
BM This is a boundary-matching method (refer to Section 9.7.2 for a detailed
description). A set of candidate vectors is $rst chosen. Each candidate
is then used to conceal the damaged block. The quality of this conceal-
ment is assessed using the side-match distortion (SMD) measure, which
is de$ned as the sum of absolute (or squared) di,erences across the four
boundaries of the block. The candidate with the minimum SMD is cho-
sen. In this section, the set of candidates includes the four neighboring
vectors—top, bottom, left, and right—and the SMD is de$ned as the SAD
across the boundaries.

MFI This is the method described in this chapter. It uses motion $eld in-
terpolation to recover one vector per pel of the damaged block. In this
section a linear interpolation kernel is employed.
Since there are four techniques in the list, there are 16 possible combina-
tions of the form S-T. Each combination leads to a di,erent long-term temporal
concealment method. For example, assume that l =(l x ;l y ;l t ), r =(r x ;r y ;r t ),
t =(t x ;t y ;t t ), and b =(b x ;b y ;b t ) are, respectively, the motion vectors of the
blocks to the left of, to the right of, above, and below the damaged block. A
combination of the form AV-BM means that the spatial components (d x ;d y )
are $rst recovered using the AV method:

ˆ
ˆ
d x = l x + r x + t x + b x and d y = l y + r y + t y + b y : (10.11)
4 4