jrhupc
commented
8 years ago Hi Team 1,
You should definitely NOT define another grid in the other frame and advance pixel by pixel. If you do define a grid then the estimated motion vectors will have a very low resolution/accuracy (they will be heavily quantized to the size of the selected blocks so it will be impossible to find small pixel movements). To avoid the big computational effort you can use a window search so only a portion of the image is searched.
Regards,
Txordi00
Hi, we are the team 1 and we have a little doubt about block matching.
In both backward and forward methodologies we have a block B in frame t or t-1 and we search its corresponding block B' in t-1 or t respectively. We assume that in forward methodology you define a grid in t-1 and in the backward case we define a grid in t frame. Regarding to the search of this B' we have this question:
Must we define another grid in the other frame (t in forward, t-1 in backward) to find B' or we must search blocks advancing pixel by pixel as seen in the attached image (backward case)? If we define a grid there is a high probability of connecting 2 different B to a single B' and we lose a little of accuracy. In the other case, we have a big computational effort and we lose the continuity within frames (or that is what we think)...