Ears In the Wild: An Efficient Ear Localisation

This research presents a novel human ear detection method that enables the utilization of the ear as a biometric trait. Classical methodologies applicable for ear recognition in controlled settings have limited resilience against many factors such as numerous ears, alterations in illumination conditions, variations in orientation, and scalability. Given the significance of detection response, a substantial increase in false positive rate can exert a considerable influence on the process of recognition. Achieving a balance between model complexity and false positive detection is a crucial aspect of designing a system that effectively minimizes the occurrence of false positive detections while maintaining an appropriate level of complexity. To tackle this issue, we provide a computationally optimized ear identification system designed to identify ears from photos and videos under uncontrolled circumstances, with the capability to operate in real-time. The suggested technique was evaluated using three databases obtained from controlled and uncontrolled situations, namely UBEAR, AWE, and UND-J2. The analyses are conducted under various and volatile settings, including a cluttered background, the presence of numerous ears, occlusions, and very small ears, in order to showcase the effectiveness and resilience of the suggested model. The obtained ear detection results demonstrate that the suggested strategy has superiority over existing contemporary ear detection techniques.