BMC Bioinformatics

Table 3 Multimodal and multitask test for noisy input

From: Deep user identification model with multiple biometric data

Modality			Noise	Accuracy (%)
ECG	Face	Finger		ID	Gender
O	O		O	94.83	95.02
O		O	O	93.68	95.21
	O	O	O	95.21	92.91
O	O	O	O	98.97	96.55
O	O			100.0	100.0
O		O		98.85	96.55
	O	O		100.0	98.85
O	O	O		100.0	99.43

From this experiment, not only does the clean modalities provides near 100% in two tasks, but in the case of noisy modalities, the feature fusion of three modalities yields 98.97% in identification. This result shows that the model architecture is robust to noises

Back to article page

ISSN: 1471-2105

Contact us

General enquiries: journalsubmissions@springernature.com