Before we start talking about our algorithm, let's talk about some of the
previous work done in this area. There has been a lot of
work on texture minification, but there has not been
as much work done to address
the artifacts during texture magnification. Here I will talk
about three important ones.
First let's look at SGI's sharpen texture extension. This extension attempts
to generate a higher resolution texture by extrapolating from levels L0 and
L1 in the mipmap. It's a simple idea, but because there is no inherent
information on the content of the texture, this technique is simply "guessing"
at what the result should be.
Another way to look at this is that there could be features so fine that they
would be lost even in the L0 level. After all, we want
to have an L0 level that is
of sufficiently low resolution so that we can pack a lot of
textures into our memory.
Thus no amount of extrapolation between
L0 and L1 will ever get those features back. For these reasons, this
extension
does not generally yield good results.