[M4IF Technotes] Unrestricted motion compensation for interlaced macroblocks

[Arvind Raman]

MPEG-4 suuports unrestricted motion compensation. In
the case of frame MBs (macoblocks), when the motion
vector points outside the frame we just pad the last
row / column of pels in the horizontal or the vertical
direction to get the predictor MB.  
Same is done for an interlaced MB when the
"horizontal" component of the motion vector points
outside the frame boundary.
According to my understanding of the standard the
difference arises when the vertical component of the
motion vector points outside the frame boundary in the
case of an Interlaced MB. 
CASE 1 : Reference field : BOTTOM FIELD
=======================================
      ^
   |--|----|
   |  |    |
   |  |    |
   |       |
   |       |
   |-------|
As shown in the above figure, if the reference field
is the BOTTOM field and the motion vector points
outside the "TOP" of the frame boundary it's the TOP
field (the first row of pels in the frame) which is
padded to from the predictor MB
CASE 2: REFERENCE FIELD : TOP FIELD
===================================
   |-------|
   |       |
   |       |
   |  |    |
   |  |    |
   |--|----|
      |
      V
Similarly if the reference field is the TOP field and
the motion vector points outside the BOTTOM of the
frame boundary it's the BOTTOM field (the last row of
pels in the frame) which if padded to form the
predictor MB.
Is my understanding of the standard correct? If so why
should this distintion be done. Wouldn't it have been
better to pad (extrapolate) the field of the same
polarity to from the predictor MB.
Thanks and regards
Arvind
__________________________________________________
Do You Yahoo!?
Yahoo! Sports - live college hoops coverage
http://sports.yahoo.com/