Table 1 Evaluation of cost function used to compute thresholds β _j when 50% of weak classifiers are missing

From: Handling missing weak classifiers in boosted cascade: application to multiview and occluded face detection

	k	Cost function	FP	TP	$\bar{n_{level}}$
			×10 ^-3
		TP_cost	3.21	0.88	1.61
P _boost	-	FP_cost	0.066	0.1	1.48
		FP_TP_cost	0.08	0.15	1.48
		TP_cost	5.56	0.95	1.29
P _knn	3	FP_cost	0.14	0.52	1.26
		FP_TP_cost	0.17	0.44	1.29
		TP_cost	5.43	0.95	1.62
P _comb	3	FP_cost	0.006	0.12	1.48
		FP_TP_cost	0.03	0.24	1.48

Three evaluation terms are exposed: the false positive rate, the true positive rate and the mean number of evaluated levels per negative example noted $\bar{n_{level}}$ .

Back to article page