[XviD-devel] help me please (about setedges() and others)

[Christoph Lampert]

On Fri, 31 Oct 2003, Nicola Bicocchi wrote:

> On Thu, 30 Oct 2003 08:08:18 +0100 (CET)
> Christoph Lampert <chl at math.uni-bonn.de> wrote:
> 
> > On Wed, 29 Oct 2003, Nicola Bicocchi wrote:
> > >   i'm playing with xvid in my thesis and i don't understand some things:
> > 
> > Hi,
> > what's the thesis about? 
> 
> A research group of my univeristy have written some code which 
> recognizes, classifies and tracks moving objects in a scene, and
> automatically recognizes background changes whenever an object
> becomes static.
> The idea is to store (or send) not the whole scene but only
> interesting objects.
> Since there are no core profile implementations avaiable, I
> decided to modify xvid to reach the goal.
> I've already written a program, but it doesn't reach
> psnr/bitrate rate values that i was attending for.
> 
> I've a couple of other questions.
> could you breifly explain me what are refH refV refHV,

Usually (in halfpel mode), noninteger (=halfpel) position are interpolated
linearly from the integer pixel positions. 
Instead of interpolating every 16x16 block which each 16x16 macroblock
might be compared with, XVID interpolated the full frame of these half-pel
position into a new frame. Since there are 3 possibilites for this: 
horizontal interpolation, vertical interpolation and interpolation in both
directions, we have 3 additional reference frame. 

So, "ref" is the pixel values at positions 
(0,0), (1,0), (2,0)
(0,1), (1,1), (2,1),
(0,2), etc.

and   refH is the pixel values at 
(0.5, 0), (1.5, 0), (2.5, 0)
(0.5, 1), (1,5, 1), ...

refV is
(0, 0.5), (1, 0.5), ...
(0, 1.5), (1, 1,5), ...

and refHV is
(0.5, 0.5), (1.5, 0.5), ...
(1.5, 0.5), (1.5, 1.5), ...


> and whait is acdc_prediction in I-frames??

The 0th DCT coefficient (so it's just the mean value of the block) is
called DC coefficient. All other DCT-coefficients are called AC.

AC and DC prediction are a way of spacial prediction in I-frames: Since
usually in neighbouring blocks DCT coefficients are similar, you can
"predict" some AC coefficients from the top neighbour or left neighbour
(you subtract them before writing the bitstream).  If that doesn't sound
promising, you can also just subtract previous DC value from present.

But if you work on MPEG-4, you should really buy a copy of the standard,
it's all in there ;-)

gruel