If you're intent is to sanitize HTML submitted by your users, I'm don't think HTML::DOM (or any other module that constructs a tree from the HTML before doing anything else) is the right route.

If not, here's what you asked:

for my $img ( $dom_tree->getElementsByTagName('img') ) {
   $img->delete();
}
[download]

Untested. Assumes getElementsByTagName returns HTML::Element objects.