Vorbis dev - Feb 2001 - A different take at video encoding - I'm stuck though

As I said in the other thread (for people using threaded mail)/in my
last mail (for the rest), I'm working on the "split up the frames into
textured triangles and output textures and coordinates" idea. The idea
is to find edges in the image. So far I have a Sobel edge detector, a
wavelet transformer (to compress the textures, still very incomplete,
transform only), and some miscellaneous stuff. The problem is, the Sobel
edge detector gives back a bunch of grey values (quick explanation, I
calc the X and Y differences between pixels, then do edge = sqrt(xd^2 +
yd^2). Dunno if it's Sobel but it looks the same as in Gimp so I suppose
it is). So what I get is a bunch of grey values one pixel smaller in
both directions as the source image.

Problem is, we need vertices and edges. What I want to do is find out
which edge points are adjacent, so a big circle counts as one edge,
which is just a bunch of points. Then triangulate that cloud of points,
using GTS (http://gts.sourceforge.net) for example.

So I find a point in the Sobel edge map that is above a certain
treshold. Each time a new point is to be considered I take the adjacent
pixels of the last pixel, and calculate a score for them. The pixel with
the highest score is taken, and another new one is added, until the
scores drop below a minimum and there's the end of an edge.

Question: How do I calculate the score? I have the direction of the
gradient at the point, the value of the Sobel edge detect, and the
coordinates of the other points in the edge.

Ideas are welcome.

Lourens

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'vorbis-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.

On Sun, 11 Feb 2001, Lourens Veen wrote:
> As I said in the other thread (for people using threaded mail)/in my
> last mail (for the rest), I'm working on the "split up the frames
into
> textured triangles and output textures and coordinates" idea. The idea
> is to find edges in the image. So far I have a Sobel edge detector, a
> wavelet transformer (to compress the textures, still very incomplete,
> transform only),
As I already mentioned before, I'd really like to see developed many
approaches. It's the same as the .gz <-> .bz2 thing, your approach
will be
considerably slower on the encoder side (on the receiver side too if
you don't have hardware OpenGL), but may achieve better compression
results. Since the 3d wavelet transform is a symmetric process, this is
probably the only approach which will allow a realtime encoder or harddisk
video recorder. But whenever compression ratio is more important than
encoding time, other algorithms may perform better.

- Holger

btw: for texture compression you can use the current wavelet code without
     major modifications -- just set the number of frames per block to 1.

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'vorbis-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.

From: "Lourens Veen" <jsr@dds.nl>
> I'm working on the "split up the frames into
> textured triangles and output textures and coordinates" idea.
How usefull would that be exactly? For fractal image coding it might build an
efficient starting poing for building a domain pool, I think thats the term, but
for motion compensation it would not be a very good one I think... shouldnt you
be more interested in segmenting along motion boundaries then?

Marco

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'vorbis-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.