If you can assume that urls will not contain (unencoded) spaces, then this might work for you:

my $test = qq|
sdfiojs pfojsdfs fs [img]sdj fpsdofj spojf sfsf [b] [img]http://www.te
+st.com/image.gif[/img] dfs fs
s fsf sfd  [img]test.gif[/img]
|;

while ($test =~ /\[img\](\S+?)\[\/img\]/g) {
   print "$1 \n";
}

__END__
c:\test>junk
http://www.test.com/image.gif
test.gif
[download]

Alternatively, maybe this is better:

while ($test =~ /\[img\]([^\[\]]+?)\[\/img\]/g) {
   print "$1 \n";
}
[download]