That sounds interesting for noticing the difference between "almost identical" versus "not", but I'd worry that the "features vary in perceived importance" part is going to resist any good calibration, since it may vary between ethnic groups and cultures of viewers.