Equal error rateEER

The point where false accepts and false rejects are equally frequent.

It gives a single number for comparing how well two face recognition algorithms perform overall, at the balance point where the two kinds of mistakes happen at the same rate.

The Detection Error Tradeoff (DET) curve plots FAR against FRR across every possible matching threshold. EER is the single point on that curve where the two rates are equal. FAR is the false acceptance rate, the share of impostors wrongly accepted, and FRR is the false rejection rate, the share of genuine people wrongly rejected.

The curve shows how the two errors trade against each other:

• At a strict threshold, FAR is low and FRR is high.
• At a loose threshold, FAR is high and FRR is low.
• At the equal error rate, the two are the same.

A lower equal error rate generally means a more accurate algorithm, since both errors can be small at the same time.

Equal error rate sits next to terms that are easy to mix up:

• False acceptance rate is the share of impostors wrongly accepted.
• False rejection rate is the share of genuine people wrongly rejected.
• Operating point is the threshold a business actually runs in production, which is rarely the equal error point.