Page 150 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience
P. 150

Section 5.2.  Warping-Based  Methods:  A  Review              127


                            Perspective:

                              u = g x (x; y)=  a 1 x + a 2 y + a 3  ;
                                          a 7 x + a 8 y +1
                                                                         (5.6)

                              v = g y  (x; y)=  a 4 x + a 5 y + a 6  :
                                          a 7 x + a 8 y +1

               Motion  estimation  and  compensation  using  higher-order  models  is  usually
            performed using the following steps:
            1.  A 2-D mesh is used to divide the current frame into nonoverlapping poly-
               gonal  patches  (or  elements).  The  points  shared  by  the  vertices  of  the
               patches are referred to as grid  or node  points.

            2.  The  motion  of  each  node  is  estimated.  This  will  map  each  node  in  the
               current  frame  to  a  corresponding  node  in  the  reference  frame.  In  e>ect,
               this will map each patch in the current frame to a corresponding patch in
               the reference frame.
            3.  For  each  patch  in  the  current  frame,  the  coordinates  of  its  vertices  and
               those  of  the  matching  patch  in  the  reference  frame  are  used  to  (nd  the
               motion parameters {a i } of the underlying motion model.
            4.  During  motion  compensation,  the  estimated  motion  parameters	{a i }  are
               substituted in the appropriate spatial transformation, Equations (5.4)–(5.6),
               to  warp  the  patch  in  the  reference  frame  to  provide  a  prediction  for  the
               corresponding patch in the current frame.
            An  example  of  this  process  is  illustrated  in  Figure  5.1.  In  this  (gure  the
            current  frame  is  divided  into  square  patches.  This  forms  a  uniform  mesh.
            During  motion  estimation,  node  points  A,  B,  C,  and  D  in  the  current  frame
                                                  �
            are mapped to node points A ,B ,C , and D in the reference frame. During
                                     �
                                        �
                                           �
                                                        �
            motion  compensation,  the  deformed  patch  A B C D is  warped  to  provide  a
                                                    �
                                                      �
                                                  �
            prediction for the square patch ABCD.
               It  should  be  pointed  out  that  there  is  a  lack  of  consistency  in  the  lit-
            erature  when  referring  to  this  type  of  motion  estimation  and  compensation
            methods.  Examples  of  the  numerous  names  employed  are  control  grid  in-
            terpolation  [108,  109,  110],  warping-based  methods  [111,  112,  113],  spatial-
            transformation-based methods [114, 115, 116, 117], geometric-transformation-
            based methods [118], generalized motion estimation methods [119, 120], and
            mesh-based methods [121, 122, 123, 124, 125, 126, 127].
               When  designing  a  warping-based  technique,  several  aspects  of  the
            method  need  to  be  considered  and  de(ned,  as  discussed  in  the  following
            subsections.
   145   146   147   148   149   150   151   152   153   154   155