[Mp4-tech] Motion estimation in H264

[mayank agarwal]

Hi all,
i am trying to understand the motion estimation in
H.264.i have following suppositions which i think are
incomplete and may be wrong.Please guide me the right
process for doing the same and any documents i can
refer to apart from H264 standard.
1.First of all we decide the block size in a
macrobloack out of 7 block sizes we choose one both in
the current frame and reference frame of a macroblock.
2.Then we compute the median of two neighbouring
pixels in each block(in half pixel) or four
pixels(full pixel)and compute the corresponding pixel
value
at each (x,y) location.
3.Then suppose in a 4*4 block we got the above values
after that we find search region in the reference
frame and calculate the SAD or SSD values for each of
the blocks.Then we find the (x0,y0) location of the
block which gives the least value of SAD.
4.Then to perform rate distortion optimization
we use the following equation
C (x, y) = SAD(x, y) + (lambda)B (x − x0, y
− y0)
and try to minimize the C(x,y).
Now i have the following doubts:
1.what is the pyramidal coding related to H.264:The
one document i am referrring to has following lines
which i could not understand:
//
In the beginning, for each MB, we compute two-level
norm pyramids for the current MB and the search
area, for 2 × 2 and 4 × 4 -pixel blocks.
//
2.Second  need to know if i am right in the
supposition that for each macroblock we need to
predict the MV from the macroblocks which are top
left,top,top right and left of the current macroblock.
3.Third i want to know more about the Lagrangian
method applied in H.264 and rate distortion
optimization.
Thanks and Regards,
Mayank Agarwal
__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com