Identity recognition from ear images is an active field of research withinthe biometric community. The ability to capture ear images from a distance andin a covert manner makes ear recognition technology an appealing choice forsurveillance and security applications as well as related application domains.In contrast to other biometric modalities, where large datasets captured inuncontrolled settings are readily available, datasets of ear images are stilllimited in size and mostly of laboratory-like quality. As a consequence, earrecognition technology has not benefited yet from advances in deep learning andconvolutional neural networks (CNNs) and is still lacking behind othermodalities that experienced significant performance gains owing to deeprecognition technology. In this paper we address this problem and aim atbuilding a CNNbased ear recognition model. We explore different strategiestowards model training with limited amounts of training data and show that byselecting an appropriate model architecture, using aggressive data augmentationand selective learning on existing (pre-trained) models, we are able to learnan effective CNN-based model using a little more than 1300 training images. Theresult of our work is the first CNN-based approach to ear recognition that isalso made publicly available to the research community. With our model we areable to improve on the rank one recognition rate of the previousstate-of-the-art by more than 25% on a challenging dataset of ear imagescaptured from the web (a.k.a. in the wild).
展开▼
