Page 273 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience
P. 273

250                Chapter 10.  Error  Concealment  Using  Motion  Field Interpolation


            within the long-term memory block-motion $eld. Note that at low frame skips,
            this  simple  ZR  method  is  suNcient,  whereas  at  high  frame  skips  the  more
            complex  method  of  BM  has  to  be  employed.  This  may  be  due  to  the  fact
            that  at  high  frame  skips,  the  zero-biased  distribution  becomes  more  spread
            (see  Property  6:3:1:2  and  Figure  6.3).  In  other  words,  d t  = 0  becomes  less
            probable,  and  longer  temporal  components  start  to  appear  more  frequently  in
            the motion $eld. Such components need to be recovered using BM. Both AV
            and MFI provide poor temporal-component recovery compared to BM and ZR.


            10.5.2  Spatial-Components Recovery
            This set of experiments investigates the best technique for recovering the spa-
            tial  components  (d x ;d y  )  of  a  damaged  long-term  motion  vector.  In  this  case,
            the  temporal  recovery  technique  T,  in  the  combination  S-T,  was  kept  con-
            stant  at  ZR,  whereas  the  spatial  recovery  technique  S  was  varied  over  ZR,
            AV,  BM,  and  MFI.  In  other  words,  four  S-T  combinations  were  considered:
            ZR-ZR, AV-ZR, BM-ZR,  and MFI-ZR.
               Figures  10.14,  10.15,  and  10.16  show  the  results  for  the  QSIF  sequences
            AKIYO,FOREMAN, and TABLE  TENNIS, respectively. Part (a) of each $gure shows
            the performance with a frame skip of 3 over a range of macroblock error rates,
            whereas part (b) shows the performance with a macroblockerror rate of 20%
            over a range of  frame skips.
               In  general,  the  best  spatial-components  recovery  is  achieved  by  MFI
            followed  by  BM.  This  is  similar  to  the  single-reference  results  reported  in
            Section  10.4.  Thus,  moving  from  a  single-reference  system  to  a  multiple-
            reference  system  does  not  signi$cantly  inOuence  the  spatial-components



                    Spatial-components recovery, Akiyo, M=10, QP=10, Skip=3   Spatial-components recovery, Akiyo, M=10, QP=10, Macroblock error rate=20%
              31                               31
                                        ZR-ZR
                                        AV-ZR   30.5
                                        BM-ZR
              30                        MFI-ZR
                                               30
              29
                                              29.5
                                               29
              PSNR Y  (dB)   28               PSNR Y  (dB)   28.5
              27
                                               28
                                              27.5
              26
                                               27
                                                                        ZR-ZR
              25
                                              26.5                      AV-ZR
                                                                        BM-ZR
                                                                        MFI-ZR
              24                               26
                10     20    30     40    50      1       2       3
      4
                         Macroblock error rate (%)          Frame skip
               (a) Performance over a range of error rates  (b) Performance over a range of frame skips
                Figure 10.14:  Spatial-components recovery for QSIF  AKIYO  with M =10 and QP =10
   268   269   270   271   272   273   274   275   276   277   278