Unless you have your image data already available as some internal format in memory, and pixel extractions are relatively infrequent, then possibly you'd be just fine with plain substr. For example, for a 8-bit non-aligned image:

sub get_pixel($$) 
{
      my ( $x, $y) = @_; 
      substr ( $image, $y * $height + $x, 1);
}
[download]

I know this is over-simplification, but I use these direct memory hacks once in a while, f.ex. $image = ~$image for images of cardinal pixels is as valid and fast as any other image negation. Your pixel extraction problem might just as well hit the same area.

Otherwise, of course, a number of libraries provide generic image access, as mentioned above.