Some people have already mentioned the 'reverse' rotation method, and
this is normaly fine if you're playing with quite large bitmaps.

However if the initial bitmaps are small and the end bitmap is to be
scaled you might want to consider super-sampling the original image and
this would also make the final image smoother.

Another interesting (though very slow) method is to try and fit a mesh of
spline curves to the RGB values of the original image.  Then if you need
to expand the image you can get the colour values of the new pixels from
this spline-mesh and so reduce aliasing without making them image look
'smudged'.

I did try this approach a few years ago, but it was in 16 bit TP6, and
was very slow.  I might be able to dig up the source if anyone's
interested.

Andy.