Page 235 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

212 Chapter 9. Error-Resilience Video Coding Techniques

information and VOP coding type, is repeated from the VOP header. This
helps error detection and resynchronization. A similar example is the picture
header repetition allowed by the optional additional supplemental enhancement
information mode (annex W) of H.263++.

9.6.2.2 Using INTRA Refresh

An e ective way to stop temporal error propagation is to periodically encode
pictures in INTRA mode. Given the large number of bits consumed by IN-
TRA pictures, this leads to a signi cant increase in the total bit rate. A more
suitable approach for applications like mobile video communication is to use
INTRA refresh on the macroblock level. By controlling the number and spatial
location of INTRA MBs, INTRA refresh can be a very e$cient and scalable
error-resilience tool.
Obviously, the required number of INTRA MBs is highly dependent on
the channel quality and capacity. Such information is usually available to
the encoder. For example, in mobile networks, antenna parameters can give
an indication of the channel quality. In Ref. 181, Haskell and Messerschmitt
discuss how to select a suitable number of INTRA MBs.
There are many methods for selecting the spatial location of INTRA MBs
within frames. One method is to choose the locations randomly [181, 182].
Another method is to follow a raster scanning order. In Ref. 183 the INTRA
MBs are placed adaptively in regions with high activity.
Recently, a very powerful technique for deciding both the number and
spatial locations of INTRA MBs has been proposed by CˆotKe et al. [182, 184].
In Ref. 182 they propose a rate-distortion optimized mode selection method for
packet lossy networks. This method takes into account the channel conditions
and the error concealment method used at the decoder. In Ref. 184 they apply
the same method to bit-oriented networks.
Obviously, if there is a feedback channel from the decoder, then information
regarding the number and locations of damaged MBs can help the encoder to
better decide the number and locations of INTRA MBs.

9.6.2.3 Using Restricted Prediction

In this technique, prediction is limited within nonoverlapping spatial
and=or temporal regions. This clearly limits temporal and=or spatial error
propagation.
For example, in the independent segment decoding mode (annex R) of
H.263+, video pictures are divided into segments. Each video picture segment
is then encoded with complete independence from all other segments in the
same picture, and also with complete independence from all data outside the
corresponding segment in the reference picture(s). For example, motion vectors