[XviD-devel] SAD vs MSE for ME (warning! acronym overflow!)

[Christoph Lampert]

On 5 Sep 2002, skal wrote:

> 
> 	Hi,
> 
> 	I found (sleep-) time to quickly write down what I
> 	was thinking of yesterday, about SAD vs. MSE.
> 	It's at:
> 
> http://skal.planet-d.net/coding/interpolation.html
> 
> 	Any opinion whether it can be useful or not?

Hi,
I just looked a little closer to do the 2D case, when I noticed
the following: 

You are constructing the optimation fractional position between 
two given integer positions. 

After ME the problem is rather to find the best fractional positions 
"around" one integer position, because ME terminates with one vector 
(and most likely the information that this is a local minimum for SAD 
when considering only fullpel). 
However, one doesn't know which of the neighbours to chose for you 
method. Do you think there is a better way than calcalating 
fraction positions for both neighbours in 1D, and for all 4 neighbours
in 2D? Because then we would have to decide which of the four is best,
so we would have to really _calculate_ SSE and check values at the four
positions :-(

gruel

But maybe you can come up with something better, because in many cases we
can assume that the four neighbours have been visited

n