Human identification using facial comparative descriptions

Eyewitness descriptions are vital for many criminal investigations, although typically still require manual discovery of possible suspects. Soft biometrics introduce a possibility to automatically search databases based on biometric features obtained from verbal descriptions. In this paper we introduce the use of comparative human descriptions for facial identification. Twenty-seven comparative traits are used to accurately describe facial features. The Elo rating system is utilized to determine continuous biometric features from multiple comparative descriptions. Experiments on the Soton gait database demonstrate a 96.7% identification accuracy with just three comparisons.