[soren at hauberg.org]

Hmm, I'm surprised that your code is slow. You could try something  
like this (untested code):

function ycbcr = rgb2ycbcr(rgb)
   A = [77, 150, 29; -44, -87, 131; 131, -110, -21]/256;
   tmp = round(A*[rgb(:,:,1)(:), rgb(:,:,2)(:), rgb(:,:,3)(:)].');
   ycbcr = cat (3, reshape (tmp (1,:), s), reshape (tmp (2,:), s),  
reshape (tmp (3,:), s));
endfunction

but I doubt it'll be much faster. Also, there is one detail about your code:

Quoting bharat pathak <bharat at arithos.com>:
>     ycbcr(:,:,1) = y;
>     ycbcr(:,:,2) = cb;
>     ycbcr(:,:,3) = cr;

if you swap the order of these commands things will be slightly faster  
because you don't have to resize 'ycbcr'.

Also, in the following part
>     y   = round((77*red  + 150*grn  + 29*blu)/256)  + 16;
>     cb  = round((-44*red - 87*grn   + 131*blu)/256) + 128;
>     cr  = round((131*red - 110*grn  - 21*blu)/256)  + 128;

Now, I can't remember my color conversions, but this looks wrong.  
Shouldn't the translations [16, 128, 128] also be divided by 256?

Søren