Page 173 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience
P. 173

150                    Chapter 6.  Multiple-Reference  motion Estimation  Techniques


                displacement of ±15 pels, SAD as the distortion measure, restricted mo-
                tion vectors, and reconstructed reference frames. Motion vectors are coded
                using the median predictor and the VLC table of the H.263 standard. The
                frame signal (in case of INTRA) and the DFD signal (in case of INTER)
                are transform encoded according to the H.263 standard. The encoder does
                not employ rate-constrained motion estimation and mode decision. Thus,
                motion  estimation  simply  chooses  the  motion  vector  that  minimizes  the
                SAD measure without any bit-rate considerations. The INTRA=INTERde-
                cision is based on heuristic thresholds and is given by the following [142]:
                      INTRA  mode is  chosen  if   A¡(SAD(d) − 500);     (6.1)

                where                  �
                                                     P
                                  A =      |f c (x; y) − B|              (6.2)
                                      (x;y)∈B
                and
                                              f c (x; y)
                                      �
                                   P
                                   B =   (x;y)∈B     ;                   (6.3)
                                            256
                where  d =(d x ;d y )  is  the  motion  vector  of  macroblock  B  in  the  current
                frame f c  and SAD(d) is the SAD between the macroblock in the current
                frame and a corresponding macroblock in the reference frame shifted by d.
            SR-RC  This  is  a  single-reference  rate-constrained  encoder.  It  is  the  same  as
                SR,  but  it  uses  rate-constrained  motion  estimation  and  mode  decision  as
                de)ned in the high-complexity mode of the H.263 test model, near-term,
                version 10 (TMN10) [142]. In this mode, motion estimation chooses the
                motion vector  that minimizes  the following  Langrangian cost  function:
                            J MOTION  = D MOTION  +   MOTION R MOTION  ;   (6.4)

                where D MOTION  is the SAD between the macroblock in the current frame
                and  the  corresponding  macroblock  in  the  reference  frame  shifted  by  d,
                R MOTION  is  the  number  of  bits  used  to  encode  the  motion  vector  d,  and
                  MOTION  is  a  Lagrange  multiplier  related  to  the  quantization  parameter
                QP using
                                     MOTION  =0:92 × QP:                 (6.5)
                To decide the mode, two Langrangian cost functions, one for each mode,
                are calculated as  follows:


                                 J INTRA  = D INTRA  +   MODE R INTRA ;   (6.6)
                                 J INTER  = D INTER  +   MODE R INTER  ;   (6.7)
   168   169   170   171   172   173   174   175   176   177   178