|
Authored by: Anonymous on Tuesday, March 26 2013 @ 04:45 PM EDT |
We have an old frame ("old") and the current frame
("current").
We are trying to reduce the amount of information needed to send the current
frame.
We do that by trying to copy pixels from the old frame to the new.
Since the objects in the old frame have moved we can't copy exactly. The motion
is represented in a vector field and we use that to map pixels from the old
frame to the new.
We look at two separate areas of the picture.
In the first area we use the vector field to map pixels from "old" to
"current" (motion_one)
In the second area we create a vector field in the same way (motion_two),
however we then try two different ways of representing the vector field:
a) we use the same field model as in the first area, but we apply a correction
field which is the difference between the motion_two and motion_one
b) we use the same vector field as in the first area but instead of correcting
the vector field we try correcting the image directly
We now do a comparison of the two methods using a cost_function which compares
the amount of data required for each method and the amount of image distortion
each method introduces.
We choose whichever one our cost function says is the best.[ Reply to This | Parent | # ]
|
|
|
|
|