I am working on a project where we have company reports with headshots pictures and names of company executives. We are examining individual differences of the executives and would like to use gender or perceived gender as a control variable. Given this is archival data, we don't have the option of collecting self-reported gender. We only have profile pictures and names available.

I'm looking for pointers to papers that have done something similar (more recently is better) or other experience shares on how this might be done while trying to minimize cultural biases of the coders. Our initial thought was to have two raters independently rate the genders and have a third rater settle any ties. Thanks in advance.