Your analysis is basically correct, notwithstanding the various tangential tuning attempts. But I did notice you could probably gain significant performance instantly using BETWEEN instead of IN, that is with most DBMS's I know, e.g.:

select photo_id from colors where ( r between 12 and 15 ) and ( g betw
+een 78 and 81 ) and b between 23 and 26
[download]

As for the data structure to load the lot into for a memory scan, there isn't anything purpose built for your problem (except possibly some linear programming modules that are OTT for your simple need). But whether hoh or aoh is the best fit depends on whether photo_id is numeric without gaps, but the lower level hash should have keys r,g,b with the respective values. Also use Storable and only update it if the database changes to avoiding having to read the whole database every time. To search the hash or array, for the aoh case something like:

my @match = map Filter( 'r', $minred, $maxred, Filter( 'g', $mingreen,
+ $maxgreen, Filter( 'b', $minblue, $maxblue, $array[ $_ ] ))), @array
+;

sub Filter {
    my ( $colour, $min, $max, $href ) = @_;
    $href or return undef();
    ( $href -> { $colour } >= $min ) or return undef();
    ( $href -> { $colour } <= $max ) or return undef();
    return $href;
}
[download]

for hoh just substitute $array[ $_ ]with $hash{ $_ } in the map statement;