Optimized Ear Biometric Recognition Using EfficientNet and PCA: A Deep Learning Approach

Ear biometric recognition is a reliable and robust method for personal identification due to the ear’s stable structural characteristics and uniqueness. This study introduces an optimized recognition system utilizing EfficientNet for deep feature extraction and Principal Component Analysis (PCA) for dimensionality reduction, improving both accuracy and computational efficiency. To address challenges such as variations in pose, lighting, and occlusion, advanced image preprocessing techniques, including Adaptive Histogram Equalization (AHE) and Gaussian filtering, are integrated. The model is trained and evaluated on multiple publicly available datasets—IIT Delhi-I, USTB Ear, AWE, AWE Extend, AMI, WPUT, UERC, EarVN1.0, and Raisoni Ear Dataset—demonstrating strong generalization across different imaging conditions. Experimental results indicate a classification accuracy of 95.7%, surpassing conventional CNN models and state-of-the-art methods. Additionally, the system improves computational efficiency, achieving an average processing time of 2.63 seconds per image. Receiver Operating Characteristic (ROC) curves and AUC values further validate the model’s robustness in distinguishing true positives from false positives. These findings highlight the potential of EfficientNet and PCA in advancing ear biometric recognition, offering a scalable and effective solution for real-world biometric security applications.