Deep user identification model with multiple biometric data

BMC Bioinformatics

Table 4 Comparison between feature-level fusion and score-level fusion

Fusion Level	Rule	Weight			Accuracy (%)
		ECG	Face	Finger	ID	Gender
Feature	-	-	-	-	98.97	96.55
Score	Sum	0.33	0.33	0.33	98.27	99.42
		0.50	0.25	0.25	98.85	99.42
		0.25	0.50	0.25	98.85	99.42
		0.25	0.25	0.50	97.70	99.42
Score	Product	0.33	0.33	0.33	96.55	89.08
		0.50	0.25	0.25	95.98	89.66
		0.25	0.50	0.25	93.10	89.08
		0.25	0.25	0.50	93.68	87.36
Score	Max	0.33	0.33	0.33	89.66	89.66
		0.50	0.25	0.25	89.66	87.93
		0.25	0.50	0.25	89.66	86.21
		0.25	0.25	0.50	89.66	87.36

From this experiment, the feature fusion of three modalities shows the best performance in user identification task. For gender classification task, the score-level fusion shows the best performance regardless of weights in each physiological data. This result shows that feature-level fusion method shows good performance even without adjusting weight

ISSN: 1471-2105