A New Approach for Large-Scale Scene Image Retrieval Based on Improved Parallel -Means Algorithm in MapReduce Environment

Table 3

Contrast of the retrieval accuracy (%) of different methods based on SUN database.

Dataset	Clustering algorithm	Precision rate (%)	Recall rate (%)	1 value
Data 9	-Means	89.29	92.16	0.907
	-Means	91.47	94.06	0.927
	Global -Means	91.91	94.87	0.934
	-Means++	92.38	94.99	0.937
	Parallel -Means	92.54	95.43	0.940
	The proposed method	96.48	98.87	0.977
Data 10	-Means	86.41	89.25	0.878
	-Means	87.96	90.81	0.894
	Global -Means	87.83	90.81	0.893
	-Means++	88.37	91.04	0.897
	Parallel -Means	91.90	94.81	0.933
	The proposed method	95.85	98.36	0.971
Data 11	-Means	81.29	84.18	0.827
	-Means	82.35	85.02	0.837
	Global -Means	83.81	86.81	0.853
	-Means++	85.26	87.94	0.866
	Parallel -Means	90.86	93.83	0.923
	The proposed method	94.93	97.89	0.964
Data 12	-Means	74.10	76.92	0.755
	-Means	76.05	78.89	0.774
	Global -Means	76.72	79.28	0.780
	-Means++	79.31	81.69	0.805
	Parallel -Means	89.53	91.90	0.907
	The proposed method	93.77	96.71	0.952
Data 13	-Means	64.83	66.43	0.656
	-Means	66.59	67.91	0.672
	Global -Means	67.28	69.05	0.682
	-Means++	69.05	72.0	0.705
	Parallel -Means	88.14	91.03	0.896
	The proposed method	92.49	95.59	0.940