Page 280 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 10.6. Discussion 257

boundary matching (BM) temporal technique. In e,ect, this improves the sec-
ond stage of temporal concealment, i.e., compensation.
Simulation results, within both an isolated error environment and an H.263
codec, showed the superior objective and subjective performances of the
designed techniques. The MFI technique achieved reasonable improvements
over conventional temporal concealment techniques, but it was found that
its performance can slightly deteriorate at very high error rates. The com-
bined BM-MFI technique showed a more superior and robust performance at
all error rates.
It was also observed that factors like spatial and temporal error propagation,
imperfections of the error detection algorithm, scene changes, and uncovered
background can severely degrade the performance of temporal concealment
techniques. Thus, despite their advantages, such techniques must be combined
with spatial techniques and must also be supported by powerful error detection
and error containment techniques.
The chapter also investigated the performance of temporal error conceal-
ment techniques when incorporated within an LTM-MCP codec. It was found
that the best techniques to recover the temporal component are zero replace-
ment (ZR) and boundary matching (BM). The former is suNcient at low frame
skips, whereas the latter is preferred at high frame skips. It was also found that
the best technique to recover the spatial components is the MFI technique. All
these $ndings were explained in view of the properties of the long-term mem-
ory block-motion $eld. In general, it was concluded that spatial-components
recovery is more complex and more important than temporal-component re-
covery. In addition, a combination of the form MFI-BM (i.e., spatial recovery
using MFI and temporal recovery using BM) will provide the best spatial-
temporal recovery. In order to achieve a more robust performance, the chapter
described the design of a multihypothesis multiple-reference temporal con-
cealment technique. In this technique, a damaged blockis concealed using
the average of four candidate concealments, probably from di,erent reference
frames. Simulation results showed the superior performance of this technique.