Comp Sci Confusion on motion estimation block diagram-MPEG video compression-:

[shivajikobardan]

Homework Statement

block matching algorithm confusion

Relevant Equations

none

My doubts are as follows-:

-> Why frame n is not segmented to blocks?

-> Why no inputs from (n-1) side for block matching?
-> What do we do in prediction error coding?

Source-: https://homepages.inf.ed.ac.uk/rbf/CVonline/LOCAL_COPIES/AV0506/s0561282.pdf

I have read this many times but that didn’t help to clear my doubts.



This video. This confused the hell out of me.

 

[Tom.G]

Homework Statement:: block matching algorithm confusion
Relevant Equations:: none

-> Why frame n is not segmented to blocks?

It is.
The poorly worded description you have only implies the block operation, for instance by showing it in the figure as "Block Matching." Later text also describes operations that can only be done on blocks.

-> Why no inputs from (n-1) side for block matching?

There is.
Again the figure shows Block Matching output for the two frames feeding the "Prediction Error Coding" operation.

Both of the above are explained (sort of) on pgs 3 and 4 of the document you linked.

-> What do we do in prediction error coding?

The Block Matching searches the blocks in the n-1 frame, and if it finds a close match to the current block in the n frame, it sends an instruction to the receiver to use that already received block from n-1.
However that n-1 block is not always an exact match due to image motion between frames.
The Prediction Error Coding then sends a 'Correction' factor that the receiver applies to n-1 frame to account for any motion that may have occurred between frames.

Note that this requires the receiver to keep a copy of the reference (previous) frame for comparison.

Hope this helps!

Cheers,
Tom

p.s. By the way, I found the video useless, I could not understand the accent of the presenter.

 

[shivajikobardan]

It is.
The poorly worded description you have only implies the block operation, for instance by showing it in the figure as "Block Matching." Later text also describes operations that can only be done on blocks.There is.
Again the figure shows Block Matching output for the two frames feeding the "Prediction Error Coding" operation.

Both of the above are explained (sort of) on pgs 3 and 4 of the document you linked.The Block Matching searches the blocks in the n-1 frame, and if it finds a close match to the current block in the n frame, it sends an instruction to the receiver to use that already received block from n-1.
However that n-1 block is not always an exact match due to image motion between frames.
The Prediction Error Coding then sends a 'Correction' factor that the receiver applies to n-1 frame to account for any motion that may have occurred between frames.

Note that this requires the receiver to keep a copy of the reference (previous) frame for comparison.

Hope this helps!

Cheers,
Tom

p.s. By the way, I found the video useless, I could not understand the accent of the presenter.

Hmm you're telling it includes everything..but it isn't included in figure I'm confused....can you explain 1 round of this flowchart in small brief? I'm really confused..btw what happens in block matching?

 

[Borek]

Very general hint: whenever I find a document on some technology confusing I google other sources to do a cross check. The more, the better.