Development of Ear Biometric Systems with Forensic Validation

Abstract

Generally, it is observed that face is the most feature dense region in human body. The natural way of identifying a human being is through looking at his or her face. But automated authentication of a user through face biometric bears some demerits like inconsistent variability due to facial expression, orientation of face, effect of aging, etc. These challenges can be overcome using ear biometric. Ear detection is easier due to its location on two sides of the head, making it orthogonal to frontal face. A person can willingly change the shape of his or her face through alteration in expression, but one cannot change his ear shape by own will. This makes ear a suitable candidate for being tested for its performance as a biometric trait. The thesis aims to extract ear feature which is rotation and scale invariant in nature, so that the feature can be used for matching two images captured in unconstrained condition. The thesis mainly concentrate on two application areas in ear biometrics: (a) Ear recognition, and (b) Forensic validation of ear biometric system. Handcrafted as well as deep features are extracted from ear images to recognize a person using ear biometric. Then forensic validation of ear recognition system is proposed, and finally forensic validation is perform with reduced feature vector size.