You need to useutf8; to tell Perl that your source file is in UTF-8. That way non-ASCII literal strings work the way you want them to.

use strict;
use warnings;
use 5.010;
use utf8;
binmode STDOUT, ':encoding(UTF-8)';

my $str = "ครัวซองเเซนด์วิชไข่ดาว Croissant Egg Sandwich ครัวซองเเซนด์วิชไข่ดาว";
$str =~ s/[^\p{Latin}\p{Common}]//g;
$str =~ s/^\s+|\s+$//g;
say $str;
__END__
Croissant Egg Sandwich

See also: Character Encodings in Perl.

Updated to unlinkify the brackets, and to exclude \p{Common} instead of \s from removal.

Thanks moritz, but when I tried this I got the output like this:

α╕äα╕úα╕▒α╕ºα&#
+9557;ïα╕¡α╕çα╣Çα╣Çα&
+#9557;ïα╕Öα╕öα╣îα╕ºα
+╕┤α╕èα╣äα╕éα╣
+êα╕öα╕▓α╕º Croissant Egg
 Sandwich α╕äα╕úα╕▒α&#9557
+;ºα╕ïα╕¡α╕çα╣Çα&#957
+1;Çα╕ïα╕Öα╕öα╣îα&#95
+57;ºα╕┤α╕èα╣äα╕é&#9
+45;╣êα╕öα╕▓α╕º
[download]

That's because it wasn't formatted correctly due to missing code tags (which were presumably left out so that the input text would be shown properly). When I first ran moritz's code, I just got the original string, but when I substituted:

$str =~ s/[^\p{Latin}\s]//g;

for this:

$str =~ s/^\p{Latin}\s//g;

it worked.

EDIT: If you have lots of extra spaces in your output, you could run it through $str =~ s/ {2,}/ /g;, too. Something to keep in mind is that moritz's approach (as is) will remove punctuation.

If what I read is actually your code, you'll need HTML::Entities and a /g modifier to s///:

$ cat test.pl
use 5.14.1;
use warnings;

use HTML::Entities;

my $image;
$image ='ครัวซองเ&#364
+8;ซนด์วิชไข&#36
+56;ดาว Croissant Egg Sandwich ครั
+วซองเเซนด&#3660
+;วิชไข่ดาว';
$image = decode_entities ($image);
$image =~ s/\p{Thai}//g;
print $image;
$ perl test.pl
 Croissant Egg Sandwich $
[download]

Why not simply use s/[^a-zA-Z ]//gr ?

You could use the transliteration operator (y/// or tr///) - see Quote-Like Operators.

$ perl -Mstrict -Mwarnings -E '
my $x = q{ครัวซองเเซนด์วิชไข่ดาว Croissant Egg Sandwich ครัวซองเเซนด์วิชไข่ดาว};
$x =~ y/ -~//cd;
say $x;
'
 Croissant Egg Sandwich