With HTML::TreeBuilder, as Your Mother already mentioned, you can do so, but please keep in mind that html may change. I have several monitors running that parse HTML constantly, and I have to change the code on a very regular basis because the people that generate or maintain the HTML keep changing it. So on true advice: be very very defensive in your parsing strategy and don't hardcode the sequence of events: the generator might add a div tag in between or swap the sequence of text and image.

use HTML::TreeBuilder;
my $tree = HTML::TreeBuilder->new;

$tree->parse_content ($html);
foreach my $img ($tree->look_down (_tag => "img")) {
    my $p = $img->parent;
    $p->tag eq "div" or next; # <img> not inside a <div>
    my $txt = $p->as_text;
    }
[download]

As you can see, this module offers you all rope you need to hang yourself or do what you need. It also offers a nice way to generate nicely formatted HTML from parsed trees:

print $tree->as_HTML (undef, "  ", {});
[download]