[vdr] Re: Commerial Filter?

[Matthias Biel <Matthias.Biel@web.de>]

Lutz Feldgen wrote:

>MB> Why do we have to know about the actual shape of the logo if we are
>MB> using motion (or a derived software) ? We could use motion to look for 
>MB> areas that are more or less constant and if we find such an area, this 
>MB> must be the logo.
>
>Here we have the problem that the logos are shiny (background is
>displayed too). So the pixel-information at these points is not really
>constant. A solution might be to compare the interesting point with
>its mask and the surrounding pixels.
>
>
Ok, but think of this algorithm:

foreach frame f
  split f vertically in a and b
  c = c + a -b

if we do this for sufficient large number of frames then c should 
contain a 'blueprint' of the logo or its negative, that
can be extracted by a threshold.
The idea behind this is that a pixel in the non logo area has another 
average value than a pixel in the logo value.

Matthias