Table 3.

Comparison of ME recognition performance composite datasets.

Method	Composite		SMIC-HS		CASME II		SAMM
	UF1	UAR	UF1	UAR	UF1	UAR	UF1	UAR
LBP-TOP (Zhao and Pietikainen, 2007)	0.588	0.578	0.200	0.528	0.702	0.742	0.395	0.410
Bi-WOOF (Liong et al., 2018)	0.629	0.622	0.572	0.582	0.780	0.802	0.521	0.512
CapsuleNet (Van Quang et al., 2019)	0.652	0.650	0.582	0.587	0.706	0.701	0.620	0.598
OFF-ApexNet (Gan et al., 2019)	0.719	0.709	0.681	0.669	0.876	0.868	0.540	0.539
Dual-Inception (Zhou et al., 2019)	0.732	0.727	0.664	0.672	0.862	0.856	0.586	0.566
STSTNet (Liong et al., 2019)	0.735	0.760	0.680	0.701	0.838	0.868	0.658	0.681
ELTRCN (Khor et al., 2018)	0.788	0.782	0.746	0.753	0.829	0.820	0.775	0.715
RCN-S (Xia et al., 2020)	0.746	0.710	0.651	0.657	0.836	0.791	0.764	0.656
STSTNet+GA (Liu et al., 2021)	0.836	0.836	0.814	0.812	0.882	0.891	0.800	0.790
RRS+CropNet(ours)	0.875	0.877	0.813	0.819	0.972	0.969	0.842	0.827
ARS+CropNet(ours)	0.911	0.904	0.855	0.851	0.974	0.979	0.912	0.893