Kernel-Based Multiview Joint Sparse Coding for Image Annotation

Table 2

Performance comparison using different methods on Corel5K, IAPRTC12, and MIRFlick25K datasets. Methods with “” means the implementation using our features. The best scores for each method are highlighted in bold. Using the same features, KMVJSC get the best results on all the three datasets. KMVJSC+T is the KMVJSC method using an additional Tag feature on MIRFlick25K dataset.

Datasets	Corel5K				IAPRTC12				MIRFlick25K
Method	AP	AR			AP	AR			AP	AR
[13]	0.26	0.32	0.287	138	0.35	0.28	0.311	252	0.32	0.24	0.274	36
[20]	0.35	0.37	0.360	160	0.45	0.32	0.374	260	0.40	0.30	0.343	38
[23]	0.29	0.33	0.309	143	0.36	0.30	0.327	256	0.33	0.24	0.278	37
KMVSC	0.35	0.38	0.364	154	0.44	0.33	0.377	262	0.39	0.29	0.333	38
MVJSC	0.34	0.36	0.350	150	0.42	0.32	0.363	261	0.37	0.29	0.325	38
KMVJSC	0.37	0.4	0.384	162	0.46	0.34	0.391	268	0.41	0.32	0.359	38
KMVJSC+T	—	—	—	—	—	—	—	—	0.43	0.33	0.373	38