Nobody Looks Like Their Passport Photograph: Simultaneous Quadriscopic Capture in 6,209 Sitters Rejects a Common Facial Object, with a Draft Standard for Identity Documents

Nobody looks like their passport photograph. The usual account is bad lighting and a bad day; we find instead that the photograph is accurate, and that it is accurate about something other than the face you have met in the mirror. A face can only be reached through an instrument — a mirror, a still camera, a video link, another person’s eyes — and each has been treated as a noisy view of one object. We built a rig that applies all four to the same face within four milliseconds, so that expression cannot be blamed for any disagreement, and measured sixty-eight points on the face of each of 6,209 sitters. Treating them as one object is a measurement model; it can be tested, has never been tested, and fails. Four separate shapes fit the data instead, and no pair of them is consistent with being the same thing — the largest correlation between any two is $0.72$ (95% CI $0.69$–$0.75$). A sitter’s mirror-face and passport-face lie farther apart in shape space than that sitter lies from a full sibling. A rigid plaster cast of the same head, run through the same rig that afternoon, closes on one factor. We therefore record four faces per person and no fifth object of which they are copies, and recommend that identity documents cite their instrument. You are the only person who has never met you.