Normally, you'd be able to use \b.

my $whitespace = qr/[\s()]+/;
my $badwords   = qr/.../i;
my $wordchar   = qr/[a-zA-Z]/;

s/ $whitespace? \b $badwords \b $whitespace? / /xg;
[download]

But since you want to allow "12345Red6789", you'll have to implement your own version of \b.

my $whitespace = qr/[\s()]+/;
my $badwords   = qr/.../i;
my $wordchar   = qr/[a-zA-Z]/;

s/
   $whitespace?
   (?<! $wordchar )  # At start of word.
   $badwords         # Words to erase.
   (?! $wordchar )   # At end of word.
   $whitespace?
/ /xg;           # Avoid joining two numbers.
[download]

By the way, Regexp::List can build an efficient $badwords.

use Regexp::List qw( );
my @badwords = qw( r rd red );
my $badwords = Regexp::List->new(modifiers=>'i')->list2re(@badwords);
# qr/r(?:e?d)?/i
[download]

Update: Added Regexp::List bit.

Thanks for the info on Regexp::List. I need to get more familiar with the modules that are out there.

I think your initial solution would end up always leaving a space where the word was? That's fine if the word was between two number but not if it was at the beginning or end. I'm happy with the cheap trim I get from removing spaces along with the bad words.

If I'm wrong about it always leaving a space, I appologize. I am committing the sin of commenting without executing the example as I don't have access to Perl on my internet connected machine.

Yes it does, in order to avoid "1234 Red 5678" becoming "12345678". Feel free to remove extra whitespace afterwards. Doing it in the regex would needlessly complicate it.

s/.../ /xg;
s/^\s+//;
s/\s+$//;
[download]

use strict;

while (<STDIN>) {
        s/[\s\(\)]+//g;         # strip optional chars
        my @matches = m/(\d+)|([^\d]+)/g;
        print join ":", grep { defined $_ and $_ !~ m/^(r|rd|red)$/i }
+ @matches;
}
[download]

$ echo "12345 (Gray) 6789 (Red)" |  ./red.pl
12345:Gray:6789
$ echo "12345 (Gray) 6789 Red" |  ./red.pl
12345:Gray:6789
$ echo "12345Gray6789Red" |  ./red.pl
12345:Gray:6789
[download]

Doesn't work?

$ echo "12345Red6789" | perl -ape 's/[\s\(]+(r|rd|red)[\s\)]+//gi;'
12345Red6789
[download]

The OP said the spaces were optional.

Updated: to handle parentheses.

#!/usr/bin/env perl

use warnings;
use strict;

while (<DATA>) {
    #my @no_reds = grep { !/^r/i } split;
    my @no_reds = grep { !/^r/i } split /[\s\(\)]/; 
    print "@no_reds\n";
}

__DATA__
 12345 Gray 6789 Red 
rd 555 blue 888 red 777 green 999
foo bar r baz ruby
goo(r)  ruby doobie
888red999
[download]

Prints out:

12345 Gray 6789
555 blue 888 777 green 999
foo bar baz
goo   doobie
888red999
[download]

Unlike ikegami's solution, this requires the input to be whitespace-separated.

Thanks for the idea. It's always good to break out of my assumption that RegEx is the right way and consider something else like Grep.

Unfortunately, this was a free-text field filled by users, so I can't assume they'll give me whitespacea, any other convention or even the results of normal human brain function.

Also, this method is either a little lacking or it's way TOO smart if it knows that rubies are actually red. :)

Update: Forget what I said, it won't work on something like "1234red5678". If I change the regex to work on this, I arrive at ikegami's solution.

I'm not sure what you want to achieve here. For the following input:

what output would you require?