Hand Detection Using Cascade of Softmax Classifiers

Table 2

The confusion matrices for detection results computed with single-resolution-based softmax detectors and multiresolution-based cascade detector. Note that row elements of matrix do not need to sum to 1 for confusion matrix of detection problem.

	Softmax+Resolution1					Softmax+Resolution2
	BK	vict	close	open	fist	BK	vict	close	open	fist
BK	0.7104	0.1480	0.1211	0.0628	0.1469	0.8205	0.0843	0.0291	0.0326	0.1145
vict	0.0420	0.9136	0.2006	0.2157	0.2585	0.0706	0.9001	0.0492	0.0856	0.1776
close	0.0272	0.5438	0.9251	0.5447	0.4443	0.0017	0.4477	0.9626	0.4077	0.4911
open	0.0532	0.3291	0.2181	0.9070	0.3531	0.0630	0.1987	0.0532	0.9055	0.2234
fist	0.0183	0.7114	0.4862	0.5146	0.9233	0.0342	0.4370	0.1935	0.2494	0.9391
	Softmax+Resolution3					The Proposed Softmax Cascade
	BK	vict	close	open	fist	BK	vict	close	open	fist
BK	0.8369	0.0889	0.0090	0.0287	0.0983	0.9884	0.0061	0.0022	0.0033	0.0053
vict	0.0761	0.8993	0.0214	0.0928	0.2292	0.0547	0.9294	0.0032	0.0389	0.0151
close	0.0026	0.4323	0.9574	0.3872	0.4655	0.0153	0.0383	0.9838	0.1319	0.0502
open	0.0862	0.1717	0.0187	0.9033	0.2219	0.0300	0.0682	0.0090	0.9700	0.0787
fist	0.0350	0.3361	0.0984	0.1435	0.9299	0.0609	0.1093	0.0459	0.0300	0.8957