CIoEBT: cancelable internet of ear biometric things based – a novel deep metric learning approach

Biometric authentication is gaining widespread acceptance globally, driving the need for robust and secure systems. This study explores the use of outer ear images as a distinctive biometric modality. The human ear, much like the face, exhibits unique and permanent features, making it a promising candidate for biometric identification. However, ear biometrics, like facial recognition, face challenges such as variations in illumination, contrast, rotation, scale, and pose. To address these issues, this paper investigates the application of Convolutional Neural Networks (CNNs), a powerful tool in computer vision, for ear recognition. Specifically, we propose a hybrid approach that combines deep CNNs with metric learning. Using CaffeNet as a feature extractor and a novel Deep Effective Pairwise Constraints Metric Learning (DEP-ML) strategy, we encode ear images into secure representations called EarCodes. These codes are further protected using the Comb-filter algorithm, resulting in highly secure and reliable biometric templates. The proposed CaffeNet-DEP-ML framework is evaluated against well-known benchmarks like VGG-verydeep16 and VGG-S on two prominent ear image datasets, AWE and USTB II. Experimental results demonstrate that our method not only outperforms current state-of-the-art techniques but also benefits from fewer trainable parameters and faster processing times. This innovative approach shows strong potential for integration into Internet of Biometric Things (IoBT) environments, offering high accuracy while ensuring privacy preservation.