Online Multikernel Learning Based on a Triple-Norm Regularizer for Semantic Image Classification

Table 1

Comparison of the results obtained with single kernel or combining-all.

(a) Caltech-101 and Caltech-256

Accuracy (%)	GB	SSIM	C-SIFT	Oriented-PDF	Combining-all
Caltech-101	81.13 ± 0.02	77.49 ± 0.06	76.45 ± 0.08	80.62 ± 0.14	88.74  ± 0.09
Caltech-256	34.33 ± 0.05	33.92 ± 0.07	33.99 ± 0.10	45.13 ± 0.15	49.64  ± 0.08

(b) Oxford Flowers (102)

Feature	Accuracy (%)
D_HSV	32.28 ± 0.16
D_HOG	39.12 ± 0.12
D_SIFTint	46.11 ± 0.44
D_SIFTbdy	27.05 ± 0.66
Combining-all	62.51 ± 0.13

(c) MNIST

Feature	Error rate (%)
SPHOG	0.72 ± 0.02
GIST	0.88 ± 0.06
LBP	15.27 ± 0.13
Combining-all	0.65 ± 0.02