Looks like it's slightly faster than the regexp:

Benchmark: timing 1000000 iterations of regexp, split, substr...
    regexp: 16 wallclock secs (16.27 usr +  0.03 sys = 16.30 CPU)
     split: 17 wallclock secs (15.88 usr +  0.00 sys = 15.88 CPU)
    substr:  6 wallclock secs ( 5.32 usr +  0.00 sys =  5.32 CPU)
[download]

Hmm, substr is the winner in a big big way. Makes sense. What if we anchor the regexp?         'regexp' => sub { my($file) = $img =~ m#.*/(.*)$# } gives us

 regexp: 17 wallclock secs (16.15 usr +  0.00 sys = 16.15 CPU)
     split: 15 wallclock secs (15.93 usr +  0.00 sys = 15.93 CPU)
    substr:  5 wallclock secs ( 5.37 usr +  0.00 sys =  5.37 CPU)
[download]

Nothing else I tried improved things.