Performance Evaluation of Some Selected Feature Extraction Algorithms in Ear Biometrics

It has been suggested by the researchers that the structural shape, size and features of the ear are unique for each person and invariant with age, which makes ear a better biometric trait, however, a major problem in ear recognition is extraction of the specific key points. This research work investigates four key feature extraction techniques Principal Component Analysis (PCA), Speeded Up Robust Features (SURF), Geometric feature extraction and Gabor filter based feature extraction techniques in terms of performance issues given by False Acceptance Rate (FAR), False Rejection Rate (FRR), Genuine Acceptance Rate (GAR) and Recognition Accuracy in order to determine the best approach (or approaches) that can best maximize security features of Ear Biometrics Systems. The results suggest the potential power of ear biometrics and demonstrate the effectiveness and efficiency of these feature extraction techniques, confirming thatPCA and Gabor feature extraction algorithms are indeed efficient and strong techniques for normal pose of the ear, obtaining Recognition Accuracies of 98.95 % and 97.93 % respectively. SURF is the most efficient in the presence of occlusion with tiny earring obtaining a GAR of 81.82 %. Gabor wavelet and SURF are invariant to rotation.