Page 57 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience
P. 57

34                                   Chapter  2.  Video  Coding:  Fundamentals



                          Analysis stage           Synthesis stage

                         h  1  n) (   ↓  2      ↑  2    g  1  n) (
                                                                   ˆ ) ( n
                 x  (n  )                                          x
                                                                 +
                         h  2  n) (   ↓  2      ↑  2    g  2  n) (

                        Analysis   Downsample   Upsample   Synthesis
                                          Coding, Channel, Decoding
                         filters                        filters
                  h  ), (  n  g  1  ) (  n   :  low-pass filters   (  n ), g  2  ) ( n   : high-pass filters
                   1                     h  2
                          Figure  2.11:  A 1-D, two-band  subband  coding system


            process  of   ltering  and  downsampling  is  called  the  analysis  stage.  The  sub-
            bands  are  then  quantized  and  coded  independently.  At  the  decoder,  the  de-
            coded  subbands  are  upsampled  (interpolated),   ltered,  and  added  together  to
            reconstruct  the  image.  This  is  knows  as  the  synthesis  stage.  Note  that  sub-
            band decomposition does not lead to any compression in itself, since the total
            number  of  samples  in  the  subbands  is  equal  to  the  number  of  samples  in
            the  original  image  (this  is  known  as  critical  decimation).  The  power  of  this
            method resides in the fact that each subband can be coded eGciently accord-
            ing  to  its  statistics  and  visual  importance.  A  block  diagram  of  a  basic  1-D,
            two-band subband  coding system is  presented in Figure  2.11.
               Ideally, the frequency responses of the low-pass and high-pass  lters should
            be nonoverlapping but contiguous and have unity gain over their bandwidths.
            In  practice,  however,   lters  are  not  ideal  and  their  responses  must  be  over-
            lapped to avoid frequency gaps. The problem with overlapping is that aliasing
            is introduced when the subbands are downsampled. A family of  lters that cir-
            cumvent this problem is the quadrature mirror  lter (QMF). In the QMF, the
             lters are designed in such a way that the aliasing introduced by the analysis
            stage is exactly cancelled by the synthesis  stage.
               The  1-D  decomposition  can  easily  be  extended  to  2-D  using  separable
             lters.  In  this  case,  1-D   lters  can  be  applied   rst  in  one  dimension  and
            then  in  the  other  dimension.  Using  a  1-D  two-band  decomposition  in  each
            direction  results  in  four  subbands:  horizontal  low=vertical  low  (LL),  horizon-
            tal  low=vertical  high  (LH),  horizontal  high=vertical  low  (HL),  and  horizon-
            tal  high=vertical  high  (HH),  as  illustrated  in  Figure  2.12(a).  This  four-band
            decomposition  can  be  continued  by  repetitively  splitting  all  subbands  (uni-
            form  decomposition)  or  just  the  LL  subband  (nonuniform  decomposition).  A
            three-stage nonuniform decomposition  is  illustrated in Figure  2.12(b).
   52   53   54   55   56   57   58   59   60   61   62