Image retrieval based on colour and improved NMI texture features

Figures & data

Figure 1. CBIR system flow diagram.

Figure 2. (a) Original (b) 2OTSU segment image (c) Optimized PCNN segment image.

Table 1. Weights of feature fusion in Corel-1k dataset.

Method	Corel-1k
	w1	w2	w3	w4
HSV	1	0	0	0
GLCM	0	1	0	0
LBP	0	0	1	0
PCNN	0	0	0	1
Improved PCNN	0	0	0	1
HSV + GLCM	0.47	0.53	0	0
HSV + GLCM + LBP	0.37	0.37	0.29	0
HSV + GLCM + PCNN	0.34	0.37	0	0.29
HSV + GLCM+ Improved PCNN	0.33	0.37	0	0.30
ours	0.26	0.29	0.22	0.23

Table 2. Weights of feature fusion in AT&T dataset.

Method	AT&T
	w1	w2	w3	w4
HSV	1	0	0	0
GLCM	0	1	0	0
LBP	0	0	1	0
PCNN	0	0	0	1
Improved PCNN	0	0	0	1
HSV + GLCM	0.46	0.54	0	0
HSV + GLCM + LBP	0.28	0.33	0.39	0
HSV + GLCM + PCNN	0.30	0.35	0	0.35
HSV + GLCM + Improved PCNN	0.29	0.34	0	0.37
Ours	0.21	0.24	0.28	0.27

Table 3. Weights of feature fusion in FD-XJ dataset.

Method	FD-XJ
	w1	w2	w3	w4
HSV	1	0	0	0
GLCM	0	1	0	0
LBP	0	0	1	0
PCNN	0	0	0	1
Improved PCNN	0	0	0	1
HSV + GLCM	0.50	0.50	0	0
HSV + GLCM + LBP	0.31	0.32	0.37	0
HSV + GLCM + PCNN	0.31	0.32	0	0.37
HSV + GLCM + Improved PCNN	0.31	0.31	0	0.38
ours	0.23	0.23	0.26	0.28

Figure 3. Precision-recall curve on Corel-1k.

Figure 4. Precision-recall curve on AT&T.

Figure 5. Precision-recall curve on FD-XJ.

Table 4. Objective evaluation indicators of system in Corel-1k dataset.

Method	Corel-1k
	Precision	Pre-recall	F-measure	MAP
HSV	0.525	0.434	0.475	0.485
GLCM	0.561	0.515	0.537	0.537
LBP	0.445	0.411	0.427	0.423
PCNN	0.436	0.353	0.390	0.414
Improved PCNN	0.451	0.449	0.450	0.438
HSV + GLCM	0.623	0.564	0.592	0.602
HSV + GLCM + LBP	0.634	0.633	0.633	0.614
HSV + GLCM + PCNN	0.662	0.547	0.599	0.651
HSV + GLCM + Improved PCNN	0.674	0.595	0.632	0.651
Ours	0.697	0.691	0.694	0.684

Table 5. Objective evaluation indicators of system in AT&T dataset.

Method	AT&T
	Precision	Pre-recall	F-measure	MAP
HSV	0.463	0.395	0.426	0.442
GLCM	0.531	0.516	0.523	0.511
LBP	0.626	0.567	0.595	0.598
PCNN	0.523	0.431	0.473	0.503
Improved PCNN	0.594	0.538	0.565	0.564
HSV + GLCM	0.644	0.613	0.628	0.621
HSV + GLCM + LBP	0.724	0.676	0.699	0.706
HSV + GLCM + PCNN	0.709	0.591	0.645	0.677
HSV + GLCM + Improved PCNN	0.715	0.629	0.669	0.682
Ours	0.746	0.709	0.727	0.723

Table 6. Objective evaluation indicators of system in FD-XJ dataset.

Method	FD-XJ
	Precision	Pre-recall	F-measure	MAP
HSV	0.406	0.278	0.330	0.379
GLCM	0.414	0.376	0.394	0.386
LBP	0.458	0.392	0.422	0.441
PCNN	0.479	0.426	0.451	0.453
Improved PCNN	0.481	0.469	0.475	0.461
HSV + GLCM	0.532	0.490	0.510	0.512
HSV + GLCM + LBP	0.547	0.6516	0.531	0.525
HSV + GLCM + PCNN	0.564	0.580	0.572	0.547
HSV + GLCM + Improved PCNN	0.596	0.518	0.554	0.579
Ours	0.629	0.618	0.618	0.603

Figure 6. (a) Query image (b) images retrieved in Corel-1k.

Figure 7. (a) Query image (b) images retrieved in AT&T.

Figure 8. (a) Query image (b) images retrieved in FD-XJ.