A Comparative Study of Classical Clustering Method and Cuckoo Search Approach for Satellite Image Clustering: Application to Water Body Extraction

Figures & data

Figure 1. Synthetic data S1 (a) and S2 (b) (from left to right).

Table 1. The characteristics of images.

Satellite	Bands	Wavelength	Acquisition date
Spot 5	Band1: Green Band2: Red Band3: Near-infrared (NIR)	0.50–0.59 µm 0.61–0.68 µm 0.78–0.89 µm	March 3, 2012
Landsat-8 Enhanced Thematic Mapper Plus (ETM+)	Band2: Green Band4: Red Band5: Near-infrared (NIR)	0.525–0.605 µm 0.63–0.69 µm 0.75–0.90 µm	May 4, 2014
Alsat-2A	Band2: Green Band3: Red Band4:Near-infrared (NIR)	0.53–0.59 µm 0.62–0.69 µm 0.76–0.89 µm	May 4, 2011

Figure 2. Real images used for experimentations.

Figure 3. Variation of the DB index with respect to the number of clusters for the S1 data set with all methods.

Figure 4. Variation of the XB index with respect to the number of clusters for the S1 data set with all methods.

Figure 5. Variation of the DB index with respect to the number of clusters for the S1 data set with all methods.

Table 2. Comparison of the optimal number of clusters obtained by the different algorithms in conjunction with the DB index, the XB index, and WB index – case of the S1 data set.

	DB		XB		WB
	K	Value	K	Value	K	Value
KM	13	0.4211	13	0.0704	20	0.2823
KHM	15	0.3963	15	0.0419	15	0.2372
FCM	15	0.3918	15	0.0417	15	0.2351
CS1	13	0.4138	16	0.0433	15	0.2336
CS2	15	0.3668	16	0.0581	15	0.2398
CSC1	15	0.3578	15	0.0341	15	0.2322
CSC2	15	0.3586	15	0.0360	15	0.2329

Figure 6. Variation of the DB index with respect to the number of clusters for the S2 data set with all methods.

Figure 7. Variation of the XB index with respect to the number of clusters for the S2 data set with all methods.

Figure 8. Variation of the WB index with respect to the number of clusters for the S2 data set with all methods.

Table 3. Comparison of the number of clusters obtained by the different algorithms in conjunction with the DB index, the XB index, and WB index – case of the S2 data set.

	DB		XB		WB
	K	Value	K	Value	K	Value
KM	15	0.5777	11	0.1061	18	0.4337
KHM	14	0.4903	14	0.0690	16	0.4148
FCM	15	0.4402	15	0.0702	15	0.3955
CS1	19	0.4623	19	0.0494	15	0.3876
CS2	17	0.4599	19	0.0643	15	0.3945
CSC1	15	0.4146	15	0.0474	15	0.3871
CSC2	15	0.4147	15	0.0498	15	0.3883

Figure 9. The histogram of the first image data set.

Table 4. Comparison of the number of clusters obtained by the different algorithms in conjunction with the DB index, the XB index, and WB index – case of the first real image.

	DB		XB		WB
	K	Value	K	Value	K	Value
KM	3	0.4971	3	0.1164	6	0.9133
KHM	3	0.5792	3	0.1429	6	0.9931
FCM	3	0.5128	3	0.1167	6	0.9117
CS1	3	0.3562	5	0.0486	5	0,8441
CS2	3	0.3562	5	0.0465	5	0.8485
CSC1	3	0.3562	5	0.0486	6	0.8321
CSC2	3	0.3562	5	0.0459	6	0.8393

Table 5. Comparison of the number of clusters obtained by the different algorithms in conjunction with the DB index, the XB index, and WB index – case of the second real image.

	DB		XB		WB
	K	Value	K	Value	K	Value
KM	3	0.3922	3	0.0704	6	0.3536
KHM	3	0.4140	3	0.0810	5	0.3640
FCM	3	0.3874	3	0.0686	5	0.3504
CS1	3	0.3198	5	0.0377	5	0.3376
CS2	3	0.3199	5	0.0373	5	0.3375
CSC1	3	0.3198	4	0.0392	5	0.3342
CSC2	3	0.3198	6	0.0381	5	0.3336

Table 6. Comparison of the number of clusters obtained by the different algorithms in conjunction with the DB index, the XB index, and WB index – case of the third real image.

	DB		XB		WB
	K	Value	K	Value	K	Value
KM	3	0.3913	3	0.0980	8	0.2966
KHM	3	0.3976	3	0.1006	8	0.3039
FCM	3	0.3818	3	0.0937	8	0.2961
CS1	4	0.3489	7	0.0470	5	0.3406
CS2	3	0.3508	3	0.0485	5	0.3400
CSC1	4	0.3488	3	0.0484	8	0.2960
CSC2	4	0.3508	3	0.0488	8	0.2972

Figure 10. Clustered images of Oran with DB index (a i), XB index, (b i), and WB index (c i) using (i = 1) KM, (i = 2) KHM, (i = 3) FCM, (i = 4) CS1, (i = 5) CS2, (i = 6) CSC1, and (i = 7) CSC2.

Figure 10. Continued.

Figure 11. The number of pixels in each cluster obtained by all the methods using WB index.

Figure 12. Water body extraction results of the first image with WB index using (a) FCM, (b) CSC1, (c) CSC2, and (d) ground truth.