Earprint recognition based on an ensemble of global and local features

Abstract

The earmarks are usual evidences in many real criminal investigations. The earprint appears for example when a criminal tries to listen through a window or a door before entering, and the methods to make it visible are similar to those used in latent fingerprint lifting. However, its acceptance as evidence in real prosecutions still raises doubts. Although it is well-accepted the uniqueness of the ear and its usefulness for person identification, the permanence of such discriminate ability in earprints is not obvious. Although the earprints do not have a powerful distinctiveness information, they are useful in a bag of evidences, being a promising soft biometric. This paper explores the discriminant properties of local descriptors for earprint-based automatic biometric recognition systems. The literature has focused on automatic systems based on the global aspect of the images, however scarcely studies have coped with in the well-known discriminate ability of earprint local characteristics. The experiments using more than 6000 images from 1200 people suggest a promising performance in comparison with previous existing proposals based on global features and encourage to further explore in this new soft biometric traits.