Robust Structure Preserving Nonnegative Matrix Factorization for Dimensionality Reduction

Table 4

Clustering performance on global noise contaminated datasets. The best performances are highlighted in bold font.

(a)

Dataset	Accuracy (%)
	means	PCA	NMF	GNMF	SPNMF	RSPNMF
FERET	23.76 ± 1.68	27.14 ± 1.57	26.9 ± 1.23	25.76 ± 0.58	28.24 ± 0.73	35.30 ± 0.81
ORL	42.36 ± 3.24	52.78 ± 2.48	52.76 ± 2.85	52.85 ± 0.79	53.91 ± 1.78	61.5 ± 1.92
Yale	36.79 ± 2.24	41.21 ± 2.98	42.03 ± 2.83	40.06 ± 1.43	43.64 ± 1.39	50.30 ± 1.57

(b)

Dataset	Normalized mutual information (%)
	means	PCA	NMF	GNMF	SPNMF	RSPNMF
FERET	50.06 ± 2.74	58.95 ± 1.02	57.94 ± 1.28	60.33 ± 0.60	62.01 ± 0.57	68.65 ± 0.43
ORL	60.14 ± 2.77	70.33 ± 1.39	67.37 ± 1.94	66.7 ± 0.80	70.99 ± 0.74	76.97 ± 0.92
Yale	41.68 ± 2.27	44.78 ± 2.09	46.94 ± 2.37	45.94 ± 0.88	47.68 ± 1.08	54.89 ± 2.08