A Novel Three Phase Approach for Single Sample Ear Recognition

Abstract

Ear recognition becomes challenging when only single training sample is available. In such scenarios, most of the existing ear recognition methods fail to work because of insufficient training data. In this paper, we propose a novel three phase approach for single sample ear recognition where in phase 1, ear images are normalized using histogram equalization, in phase 2, a novel ear representation called Two Dimensional (2D) Eigenears is proposed and in the last phase, classification is carried out using nearest neighbour classifier. The proposed approach is the first approach in single sample ear recognition which is based only on two dimensional (2D) ear images. Here, normalized ear image of each identity is represented as a linear combination of the so called 2D Eigenears thereby reducing the time complexity of the proposed approach. Experimental results on two publicly available ear datasets viz. IIT Delhi and CP show the efficacy of the proposed approach. The proposed method achieves more than 70% and approximately 80% rank-1 recognition accuracy on IIT Delhi and CP databases respectively.