Table 2.

A comparison of rank-one (R1), rank-five (R5), and area under cumulative match characteristic (AUC) for the different VGG-based models on the AMI, its cropped version AMIC, and WPUT ear datasets. The results are given in percentages and the top three metrics for each learning method on each dataset are highlighted in bold. The reported results from our ensembles on the AMI and WPUT datasets indicate significant improvements of recognition rates over recent studies in the literature.

Methods	Model	AMI			AMIC			WPUT
		R1	R5	AUC	R1	R5	AUC	R1	R5	AUC
Scratch Training	VGG-11	83.21	$\mathbf{97.14}$	$\mathbf{98.57}$	69.28	87.14	96.61	$\mathbf{47.83}$	$\mathbf{71.68}$	$\mathbf{96.40}$
	VGG-13	84.28	96.07	98.51	$\mathbf{70.35}$	$\mathbf{87.86}$	$\mathbf{96.70}$	46.56	69.26	96.24
	VGG-16	$\mathbf{86.07}$	96.79	98.54	$\mathbf{70.35}$	84.29	96.34	46.17	68.75	96.20
	VGG-19	84.28	95.00	98.45	68.92	83.21	95.65	44.13	66.07	95.96
Feature Extraction	VGG-11	76.07	93.92	98.24	$\mathbf{80.35}$	$\mathbf{93.21}$	$\mathbf{97.96}$	$\mathbf{59.44}$	$\mathbf{81.12}$	$\mathbf{98.19}$
	VGG-13	81.78	$\mathbf{97.50}$	$\mathbf{98.61}$	71.78	89.64	97.01	51.53	76.15	97.56
	VGG-16	$\mathbf{83.21}$	93.92	98.22	74.64	88.21	96.46	50.13	73.72	97.54
	VGG-19	68.92	91.42	97.53	62.85	81.79	95.76	44.26	69.52	97.10
Fine-Tuning	VGG-11	94.29	98.93	98.83	89.64	95.36	98.28	71.05	86.61	98.56
	VGG-13	94.64	98.57	98.86	90.36	96.79	$\mathbf{98.67}$	73.98	88.78	$\mathbf{98.71}$
	VGG-16	96.07	$\mathbf{99.29}$	98.87	89.29	97.50	98.42	73.98	$\mathbf{89.16}$	98.70
	VGG-19	$\mathbf{96.78}$	$\mathbf{99.29}$	$\mathbf{98.92}$	$\mathbf{92.14}$	$\mathbf{97.86}$	98.60	$\mathbf{74.36}$	88.65	98.69
Our Ensembles	VGG-13-16	97.14	99.64	98.91	92.85	96.42	98.57	76.40	90.69	98.87
	VGG-13-16-19	$\mathbf{97.50}$	99.64	98.41	92.85	$\mathbf{97.85}$	$\mathbf{98.67}$	78.19	$\mathbf{91.07}$	98.89
	VGG-11-13-16-19	97.14	99.64	98.91	$\mathbf{93.21}$	96.78	98.63	$\mathbf{79.08}$	90.43	$\mathbf{98.92}$
Previous Work	Chowdhury et al. [69]	67.26	-	-	-	-	-	65.15	-	-
		63.53	-	-	-	-	-	67.00	-	-
	Hassaballah et al. [9]	72.86	-	-	-	-	-	38.76	-	-
		73.71	-	-	-	-	-	37.13	-	-
	Alshazly et al. [10]	70.20	-	-	-	-	-	-	-	-
	Raghavendra et al. [70]	86.36	-	-	-	-	-	-	-	-
	Sultana et al. [71]	-	-	-	-	-	-	73.00	86.00	-