A Novel Deep Neural Network Model using Resnet50-BiLSTM for Biometric Ear Recognition in Degraded Conditions

Human identification systems that rely on biometrics are highly sought after as security and privacy concerns continue to rise. The ear biometric, being distinctive and convenient for identification purposes, offers several advantages over other popular biometrics like the face, palmprint,fingerprint and iris. Considerable effort has been dedicated to ear biometrics. Besides, the current techniques have achieved impressive success when dealing with limited databases. However, when faced with an unconstrained environment, the recognition process becomes significantly more challenging due to the various obstacles encountered by the captured images. This paper puts forward a novel approach for ear recognition in degraded conditions based on Resnet50 and Bidirectional Long Short Term Memory (BiLSTM) models. Our proposed approach recorded the best accuracy of 96.07% benchmarking against existing techniques with the EVDDC database.