Ear Recognition based on Local Texture Descriptors

Abstract

Automated personal identification using unique characteristics of the human ear is emerging as an appealing modality in forensic and biometric domains. This study investigates the use of local texture descriptors for ear recognition, and the effects of the fusions of these descriptors. The study presents an investigation of Local Binary Patterns (LBP) and provides extensions of various local descriptors, namely Local Ternary Patterns (LTP), Local Directional Pattern (LDP) and Directional Ternary Patterns (DTP) to ear recognition. A novel approach is proposed for the fusion of these descriptors, referred to as Fusions of Local Descriptors (FLD). Experiments were performed on the publicly available IIT Delhi databases (IITD-l and IITD-2), consisting of several subjects under varying conditions. The experiments show exceptional results, highly competitive with and in some cases beyond the state-of-the-art. Best recognition rates yielded from the FLD fusing DTP and multi-resolution LBP. This study achieved a recognition rate of 95.88% and 97.44% on IITD-l and IITD-2 respectively.