Page 168 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience
P. 168

Section 6.3.  Long-Term  Memory Motion-Compensated Prediction   145


            frame  in  memory.  This  sliding-window  technique  will  be  adopted  throughout
            this  chapter.


            6.3.1  Properties of Long-Term Block-Motion Fields
            This  subsection  investigates  the  properties  of  long-term  block-motion  )elds
            and  compares  them  to  those  of  single-reference  block-motion  )elds.  All  il-
            lustrations  in  this  subsection  were  generated  using  a  full-pel  full-search  long-
            term memory block-matching algorithm applied to the luma component of the
            FOREMAN  sequence with blocks of 16 × 16 pels, a maximum allowed displace-
            ment  of  ±15  pels,  SAD  as  the  distortion  measure,  restricted  motion  vectors,
            and original  reference frames.

            Property 6.3.1.1  The  distribution  of  the  long-term  memory  spatial  displace-
            ments (d x ;d y  ) is center-biased. This is evident from Figure 6.2, which shows
            the distribution of the relative frequency of occurrence of the spatial displace-
            ments  d x  (Figure  6.2(a))  and  d y  (Figure  6.2(b)).  Note  that  this  is  similar  to
            the  single-reference  case  (M =1; skip = 1),  although  in  the  case  of  multiple-
            reference  (M =50; skip = 1),  the  distribution  is  slightly  more  spread,  which
            indicates  that  longer  displacements  are  slightly  more  probable.  This  distribu-
            tion  is  even more spread  at  higher frame skips, (M =50; skip = 4).

            Property 6.3.1.2  The  distribution  of  the  long-term  memory  temporal  dis-
            placement  d t  is  zero-biased.  This  is  evident  from  Figure  6.3,  where  the  tem-
            poral  displacement  d t  = 0  (which  refers  to  the  most  recent  reference  frame



                           QSIF Foreman                    QSIF Foreman
               0.6                            0.7
                                      M=1, Skip=1                     M=1, Skip=1
                                      M=50, Skip=1                    M=50, Skip=1
                                      M=50, Skip=4                    M=50, Skip=4
                                              0.6
               0.5
                                              0.5
               0.4
              p(d x )   0.3                   p(d y )  0.4
                                              0.3
               0.2
                                              0.2
               0.1
                                              0.1
               0                               0
               −15   −10   −5   0   5   10  15   −15   −10   −5   0   5   10  15
                              d x                            d  y
                (a) Distribution  of relative frequency of   (b) Distribution  of relative frequency of
                occurrence of d x              occurrence of d y
             Figure 6.2:  Center-biased distribution of  the long-term memory  spatial  displacements (d x;d y  )
   163   164   165   166   167   168   169   170   171   172   173