Page 245 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience
P. 245

222                         Chapter 9.  Error-Resilience  Video Coding  Techniques



                          Concealment              Displacement
                          Displacement
                                                   Compensation
                           Estimation
                                  (a) Stages of temporal concealment





                         displacement   displacement compensation



                                                           damaged
                                                            block




                          reference frame               current frame
                                (b) Conventional temporal concealment
                               Figure 9.8:  Temporal error concealment


            i.e., displacement estimation, is bypassed and the concealment displacement is
            simply set to the original  motion vector.
               In practice, however, the motion vector of a damaged block is usually lost
            or  erroneously  received.  This  is  due  mainly  to  spatial  error  propagation.  For
            example, an erroneous codeword will usually lead to loss of synchronization at
            the decoder and all blocks, including their motion information, up to the next
                                                                3
            synchronization point will be undecodable and completely lost. In such cases,
            the  displacement  estimation  stage  at  the  decoder  is  extremely  important.  In
            fact, the only di erence between the various conventional temporal techniques
            reported  in  the  literature  is  in  their  displacement  estimation  algorithm.  This
            stage  is  also  known  as  motion  information  recovery,  because  it  attempts  to
            recover or  provide  an approximation  to the original  motion information.
               The  simplest  and  most  commonly  used  technique  is  to  replace  the  dam-
            aged motion vector with (0; 0) [179, 192]. This is based on the center-biased
            property of video block-motion  elds, which is also equivalent to the temporal
            smoothness  property  of  video  signals.  The  technique  is  usually  referred  to  as


              3 As already discussed, RVLCs and data partitioning into motion and texture data are some of
            the mechanisms that can be used  to reduce this  e ect.
   240   241   242   243   244   245   246   247   248   249   250