Page 93 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

70 Chapter 3. Video Coding: Standards

3.4.7 H.263, Version 3 (H.263++)
Version 3 of the H.263 standard is informally known as H.263++. This
version adds a number of optional feature enhancements to versions 1
and 2.

3.4.7.1 Enhanced Reference Picture Selection Mode (Annex U)
The enhanced reference picture selection (ERPS) mode is an enhancement to
the RPS mode (annex N) of H.263+. In addition to enhancing error resilience,
this mode provides bene/ts in terms of coding e,ciency.
As with the RPS mode, the ERPS mode extends the motion estimation and
compensation processes to use more than one reference picture. In the ERPS
mode, however, enhanced performance is achieved by allowing reference pic-
ture selection on the macroblock, rather than the picture, level. Thus, in this
case, each motion vector is extended by a picture reference parameter that is
used to address a macroblock or block prediction region in any of the multiple
reference pictures.
The ERPS mode also includes a submode for improving the coding
e,ciency of B-pictures. In this submode, encoders can use more than one
reference picture for both forward and backward prediction of B-pictures.
Another submode of ERPS is provided to reduce memory requirements.
In this submode, each reference picture is partitioned into smaller rectangular
units called subpictures. The encoder can then indicate to the decoder that
speci/c subpicture areas of speci/c reference pictures will not be used as a
reference for the prediction of subsequent pictures. This allows the memory
allocated in the decoder for storing these areas to be used to store data from
other reference pictures.

3.4.7.2 Data Partitioned Slice Mode (Annex V)
In this mode, data is arranged in a video picture segment as de/ned in the
independent segment decoding mode (annex R) of H.263+. The contents of
this segment are rearranged such that the header information for all the MBs
in the segment are encoded and transmitted together, followed by the motion
vectors for all the MBs in the segment and then by the DCT coe,cients for all
the MBs in the segment. The segment header uses the same syntax as the slice
structured mode (annex K) of H.263+. The header, motion vectors, and DCT
partitions are separated by markers. In addition to data partitioning, this mode
uses RVLC tables for encoding header and motion information. As will be
discussed later, data partitioning and RVLC provide robustness in error-prone
environments. Another error-resilience enhancement in this mode is that the
motion vector predictor is no longer formed from three neighboring motion