Ear Biometric Recognition Using Gradient-Based Feature Descriptors

Recently, intensive research efforts are conducted on the human ear as a promising biometric modality for identity recognition. However, one of the main challenges facing ear recognition systems is to find robust representation for the image information that is invariant to different imaging variations. Recent studies indicate that using the distribution of local intensity gradients or edge directions can better discriminate the shape and appearance of objects. Moreover, gradient-based features are robust to global and local intensity variations as well as noise and geometric transformation of images. This paper presents an ear biometric recognition approach based on the gradient-based features. To this end, four local feature extractors are investigated, namely: Histogram of Oriented Gradients (HOG), Weber Local Descriptor (WLD), Local Directional Patterns (LDP), and Local Optimal Oriented Patterns (LOOP). Extensive experiments are conducted for both identification and verification using the publicly available IIT Delhi-I, IIT Delhi-II, and AMI ear databases. The obtained results are encouraging, where the LOOP features excel in all cases achieving recognition rates of approximately 97%.