Page 103 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience
P. 103

80                                      Chapter 3.  Video Coding:  Standards


            based  motion  estimation  and  compensation  are  adapted  to  the  arbitrary-shape
            VOP  structure  of  MPEG-4.  The  standard  has  three  modes  for  encoding  a
            given VOP: intra-VOP (I-VOP), predicted-VOP (P-VOP), and bidirectionally-
            predicted-VOP (B-VOP).
               Since the shape, size, and location of a VOP can change from one instance
            to  another,  the  absolute  (frame)  coordinate  system  is  used  for  referencing
            every  VOP.  Thus,  the  motion  vector  for  a  particular  feature  inside  a  VOP
            refers to the displacement of the feature in absolute coordinates. During motion
            estimation and compensation, no alignment of VOP bounding boxes at di1erent
            time instances is performed.
               Motion  is  estimated  only  for  those  MBs  within  the  bounding  box  of  the
            current  VOP.  If  the  current  MB  is  an  internal  MB,  then  motion  is  estimated
            using  the  usual  block-matching  method.  If,  however,  the  current  MB  is  a
            boundary  MB,  then  motion  is  estimated  using  a  modi/ed  block-matching
            method called polygon matching. In polygon matching, the distortion measure
            is  calculated  using  only  those  pels  in  the  current  macroblock  that  belong  to
            the VOP.
               The  motion  estimation  and  compensation  processes  may  require  accessing
            pels  outside  the  reference  VOP.  Padding  is  used  to  de/ne  the  values  of  such
            pels.  The  luma  component  is  padded  per  16 × 16  samples,  while  the  chroma
            components  are  padded  per  8 × 8  samples.  If  the  reference  MB  is  a  bound-
            ary  MB,  then  it  is  padded  using  repetitive  padding.  This  process  starts  by
            horizontal  repetitive  padding,  where  each  sample  at  the  boundary  of  a  refer-
            ence  VOP  is  replicated  horizontally  in  the  left  and=or  right  direction  in  order
            to  /ll  the  transparent  region  of  the  reference  MB.  If  there  are  two  boundary
            sample  values  for  /lling  a  sample,  the  two  boundary  samples  are  averaged.
            The  remaining  un/lled  transparent  samples  are  padded  by  a  similar  process
            as  the  horizontal  repetitive  padding  but  in  the  vertical  direction,  i.e.,  vertical
            repetitive padding. The remaining MBs within the reference VOP are exterior
            MBs.  Such  MBs  are  /lled  by  extended  padding.  In  this  method,  samples  of
            an exterior MB are /lled by replicating the samples at the border of the neigh-
            boring  boundary  MB.  If  an  exterior  MB  is  next  to  more  than  one  boundary
            MB, then one of the boundary MBs is chosen according to a priority criterion
            de/ned  by  the  standard.  The  remaining  exterior  MBs  are  /lled  with  128  (for
            an 8-bit luma component).
               Motion  vectors  are  estimated  to  half-pel  accuracy.  They  are  then  predic-
            tively VLC coded in a similar fashion to the H.263 standard.
               Similar  to  the  H.263  standard,  MPEG-4  has  an  advanced  prediction  mode
            (four  motion  vectors  per  MB  and  unrestricted  motion  vectors)  and  an  over-
            lapped motion compensation  mode.
   98   99   100   101   102   103   104   105   106   107   108