Multimodal Markov Random Field for Image Reranking Based on Relevance Feedback

Table 2

Relative improvements (%) in terms of MAP of relevance feedback alone (RF), textual MRF (t-MRF), visual MRF (v-MRF), multimodal MRF (MM-MRF), the two intermedia relevance feedback variants (IRF (T > V) and IRF (V > T)), and the late fusion (LF) method over the base image retrieval system.

𝑘	RF	t-MRF	v-MRF	MM-MRF	IRF; T > V	IRF; V > T	LF
1	5.38	20.90	8.90	27.34	24.16	23.31	16.53
3	16.24	32.39	18.48	33.70	34.29	34.17	27.64
5	26.49	41.75	27.85	45.49	42.81	42.48	36.12
8	42.09	59.05	43.03	63.25	59.60	59.52	52.40
10	50.91	66.13	51.76	70.88	66.77	66.47	59.81