| Awadhesh 2006-12-11, 6:57 pm |
| Hi all,
I am working on the Whole frame loss error concealment technique. I am
going through the proposed algorithm for "Frame Concealment for
H.264/AVC Decoders" by Pierpaolo Baccichet. I have some doubts
regarding calculation of Nij for the macroblock.
Let me explain my question with the following example so that I can
make you understand my doubts clearly.
Suppose the following 2x2 matrix represents our block.
(4,2) (3,2)
(1,0) (0,0)
Then,
mean_x = (4+3+1+0) / 4 = 2.
mean_y = (2+2+0+1) / 4 = 1.2 = 1.
variance_x = ((4-2)^2 + (3-2)^ + (1-2)^2 + (0-2)^2) / 4 = 2.5
= 2
variance_y = ((2-1.2)^2 + (2-1.2)^2 + (0-1.2)^2 + (0-1.2)^2) /
4 = 1.3 =1
1) What is the meaning of "Number (Nij) of MV's falling in to 16x16
macroblock?
I think that Nij represents the number of pixels in the
macroblock having individual motion vectors less than the mean value
for the corresponding macroblock. So, In this case:
Nij_x = 2.
Nij_y = 2.
Is this interpretation correct?
2) I have used macroblock of fixed size 16x16. So for all the 256
pixel, I am using the motion vector of the corresponding macroblock.
Hence the mean value for projection is same as the motion vector of the
matroblock itself. In the above example, suppose the motion vector for
the block is (1,0). Then all the four pixels will have their individual
motion vectors as (1,0).
So, mean_x = 1.
mean_y = 0.
variance_x = 0.
variance_y = 0.
In this case can anyone suggest me how can I implement the 4x4 based
field estimation because for all vaues of the motion vectors of the
macroblock, I am getting mean value same as motion vector for that
macroblock and zero variance. Hence I am unable to decide when to go
for block level motion field estimation.
Regards,
Awadhesh.
|