If you had a page that followed the W3C standards, couldn't you grab the page with LWP::Simple, store that to a file, get the size, and use HTML::TokeParser to parse out the image sizes?

Update: After thinking about this on a long drive today ( and checking the W3C website to make sure ) I realized that all you're going to be able to parse are the alloted pixel sizes of a web page. I beleive that other monks in the thread have pointed out ways to get image sizes.