Like any system, also image recognition systems are prone to errors. These errors are normally described by using two related indexes: false positives (false acceptance rate or FAR) and false negatives (false rejection rate or FRR). FAR represents the percentage of cases in which the system incorrectly identifies an object that is not present in database, confusing it with one of those present, while FRR represents the percentage of cases in which the system does not identify an item which actually exists in the database.
These two values are usually inversely related, and vary depending on the configuration parameters and the recognition threshold adopted. If the system behavior is characterized by a perfect balance (equal percentages) of FAR and FRR, this value is named equal error rate or EER .