theora - Aug 2006 - Simple RGB->YUV Stuff...

I've spent the weekend or so trying to get my head around Theora and I'm
just not having any luck. First of all, I really don't have any interest
in immersing myself in video encoding--I just have one simple thing I'd
like to do--so I haven't spent any time reading about the theory or
specs or anything like that.

I'm attempting to write a "render-to-theora" target for OpenGL.
Basically, I have access to a buffer of RGBA data that I want to convert
into YUV and pass to Theora to encode for me.

There are lots of different ways to do the per-pixel conversion from RGB
to YUV that I've found online; I think I have this part down fairly
well. Eventually, what I end up with are three buffers, each one as
large as the other and containing the from-RGB-to-YUV values performed
per-pixel. This is where I get hung up.

To pass the YUV data to the various encode functions I have to create a
yuv_buffer struct and set the widths, heights, and strides correctly.
What I've been trying to use is:

        yuv_buffer yuv;

        yuv.y_width   = Width;
        yuv.y_height  = Height;
        yuv.y_stride  = Width;
        yuv.uv_width  = Width;
        yuv.uv_height = Height;
        yuv.uv_stride = Width;
        yuv.y         = YDATA;
        yuv.u         = UDATA;
        yuv.v         = VDATA;

There isn't a lot of documentation on how Theora "wants" the pixel
data
in theora.h, so I really feel like this should be right (since I have a
1:1:1 ratio of RGB data to YUV data). The Y data is perfect--that is,
the grayscale image in my video is perfect. However, the UV stuff is
very wrong, and I've included a screenshot here:

http://cherustone.com/theora.png

In that screeny, the "blue" bars should be right on top of the
grayscale
beneath it (the real image is just a left-to-right blue line drawn by
Cairo).

If anyone has any advice I would very, VERY much appreciate it. It seems
like a simple enough thing to do, and I imagine lots of other people
would be interested in it. I don't really think it's a question of
"how
do I convert from RGB to YUV", but rather "how do I populate the
yuv_buffer struct properly so libtheora interprets my data correctly."

Am I missing something very fundamental here? :)

On 8/21/06, Jeremy Moles <jeremy@emperorlinux.com> wrote:
> There are lots of different ways to do the per-pixel conversion from RGB
> to YUV that I've found online; I think I have this part down fairly
> well. Eventually, what I end up with are three buffers, each one as
> large as the other and containing the from-RGB-to-YUV values performed
> per-pixel. This is where I get hung up.
>
> To pass the YUV data to the various encode functions I have to create a
> yuv_buffer struct and set the widths, heights, and strides correctly.
> What I've been trying to use is:
>
>         yuv_buffer yuv;
>
>         yuv.y_width   = Width;
>         yuv.y_height  = Height;
>         yuv.y_stride  = Width;
>         yuv.uv_width  = Width;
>         yuv.uv_height = Height;
>         yuv.uv_stride = Width;
>         yuv.y         = YDATA;
>         yuv.u         = UDATA;
>         yuv.v         = VDATA;
Theora support several different pixel formats, but the libtheora
encoder only supports a single one of those. It's commonly called
"4:2:0", a notation which means that there's chroma subsampling
both
horizontally and vertically.

What that means is that the two chroma planes ('u' and 'v')
should
have half the width, and half the height, of the luminance ('y')
plane. You'll want to filter your existing data to do that.

Mike