Landmark-Based Facial Feature Construction and Action Unit Intensity Prediction

Table 3

Result for the AU intensity estimation.

AUs	MSE				CORR
	Proposed	RBNN	BPNN	SVR	Proposed	RBNN	BPNN	SVR
AU01	0.024178	0.026	0.047	0.051	0.992	0.991	0.957	0.960
AU02	0.009237	0.012	0.013	0.027	0.997	0.997	0.970	0.978
AU04	0.009893	0.012	0.066	0.059	0.995	0.995	0.893	0.824
AU05	0.027837	0.030	0.134	0.123	0.992	0.991	0.881	0.895
AU06	0.022889	0.025	0.119	0.107	0.989	0.989	0.841	0.859
AU07	0.012078	0.013	—	—	0.994	0.994	—	—
AU09	0.005872	0.010	—	—	0.998	0.998	—	—
AU10	0.024854	0.024	0.036	0.033	0.996	0.988	0.924	0.939
AU12	0.038977	0.039	0.042	0.039	0.985	0.985	0.939	0.930
AU14	0.016743	0.021	—	—	0.994	0.988	—	—
AU15	0.038062	0.042	0.056	0.060	0.981	0.979	0.890	0.892
AU16	0.015238	0.017	—	—	0.993	0.992	—	—
AU17	0.021056	0.023	0.043	0.041	0.990	0.989	0.896	0.923
AU18	0.018616	0.019	0.064	0.056	0.992	0.992	0.955	0.947
AU20	0.020243	0.025	0.058	0.046	0.987	0.983	0.878	0.913
AU22	0.009127	0.018	—	—	0.997	0.997	—	—
AU23	0.025397	0.028	0.092	0.099	0.982	0.980	0.921	0.925
AU24	0.009601	0.012	0.149	0.126	0.996	0.995	0.790	0.863
AU25	0.009886	0.011	—	—	0.995	0.994	—	—
AU26	0.017167	0.018	0.025	0.031	0.988	0.988	0.954	0.976
AU27	0.008465	0.008	0.097	0.104	0.996	0.996	0.931	0.969
AU28	0.016481	0.024	—	—	0.994	0.991	—	—

AU, action unit; BPNN, backpropagation neural network; CORR, correlation coefficient; MSE, mean square error; RBNN, radial basis neural network.